Extreme environments are unstructured and change rapidly, making human exploration in unfamiliar areas difficult. Construction robotics can help reduce risks to human safety and property in these environments by integrating digital technology and artificial intelligence. This technology has the potential to significantly improve the quality and efficiency of construction, making it a key area for future research. Extreme environments include hazardous work sites, polluted areas, and harsh natural conditions. Our review of construction robotics in these settings highlights several knowledge gaps. We focused on four main areas: mechanism design, perception, planning, and control. Our analysis reveals challenges in practical applications, such as creating adaptable mechanisms, accurately perceiving changing environments, planning for unstructured sites, and optimizing control models. Future research should explore: biomimetic designs inspired by nature, multimodal data fusion for perception, adaptive planning strategies, and hybrid control models that combine data-driven and mechanism-based approaches.
Ke You, Cheng Zhou, Lieyun Ding, Yuxiang Wang.
Construction Robotics in Extreme Environments: From Earth to Space.
Engineering, 2025, 55(12): 107-124 DOI:10.1016/j.eng.2024.11.037
Construction in extreme environments shows human ingenuity and the exploration of uncharted territories. However, the diverse and unpredictable nature of these environments makes construction complex and dangerous. This poses safety risks to construction workers, especially in hazardous, polluted, and harmful settings such as deep earth, deep sea, and deep space [1], [2], [3]. While construction robots have proven effective in typical construction scenarios, addressing technical and scientific challenges becomes essential when they operate in extreme environments. These challenges include modeling and optimizing construction robots in environments with high-dimensional complexity and uncertainty [4], [5], [6]. Construction robots can replace workers in repetitive tasks, improving efficiency and addressing labor shortages [7], [8]. However, their application in extreme environments requires prompt attention to overcome these challenges.
Extreme environments often exhibit unstructured and dynamic characteristics, rendering the design and application of construction robots exceedingly complex. Construction robots are acknowledged for their effectiveness and adaptability in unstructured construction sites, underscoring the considerable potential of robotics in this domain [9], [10]. The purpose of this article is to thoroughly examine the technological limitations and knowledge gaps that construction robots face when operating in extreme environments. It is crucial to have a comprehensive understanding of these challenges in order to guide future research and technological advancements. Given the unpredictable nature of construction sites, the ability of robots to perceive and adapt to their environment is of utmost importance. Addressing this issue effectively requires extensive research and technological advancements that can cater to the demands of various extreme environments [11], [12].
The application of construction robots in extreme environments can be assessed from four dimensions: ➀ mechanism design, ➁ perception, ➂ planning, and ➃ control, as shown in Fig. 1 [13], [14], [15], [16]. Mechanism design plays a crucial role in how construction robots interact with their environment. By designing mechanisms that minimize resistance, construction robots can navigate complex terrain swiftly and stably. Precise environmental perception is essential for ensuring the safety of both humans and assets during construction. However, in extreme environments, robot perception accuracy is limited by various factors. The fusion of multimodal data for perception is key to enhancing the autonomy of construction robot systems [1], [11], [17]. Motion planning is another important aspect for construction robots operating in extreme environments. These robots must be able to plan their movements based on the surrounding environment, taking into account the dynamic and uncertain nature of the context. They must also consider various constraints, including environmental and temporal factors, as well as the inherent dynamics constraints of the robots themselves [18], [19]. The hybrid approach of combining data-driven and mechanism-driven control offers innovative and practical solutions to addressing the challenges posed by extreme environments [2], [20], [21].
Extensive research into robot kinematic modeling and bionic design has paved the way for the application of robots in extreme conditions [22]. Construction robots are engineered with specialized structural designs to meet the unique demands of their operating environments. In extreme environments, robots equipped with specialized structures exhibit exceptional performance and have the ability to adapt their posture, allowing them to handle various tasks and work scenarios effectively [12], [23], [24], [25]. Intelligent decision-making plays a crucial role in addressing complex and unpredictable construction scenarios [1], [26]. Continued exploration and innovation in this field will enable robots to adapt to and operate autonomously in complex and extreme construction environments. This article, based on a comprehensive scientific analysis of current research, identifies potential avenues for future development and expansion of the field [27]. It aims to provide valuable guidance to advance research and application in the domain of construction robots, with the goal of optimizing their potential in various application fields to meet the challenges of extreme environments.
This article systematically examines and analyzes the domains of mechanism design, perception, planning, and control of construction robots. The structure of the paper is as follows: Section 2 presents a collaborative analysis of key technologies in construction robots, utilizing keyword-based and cluster-based topic analyses to identify their applications. Section 3 investigates the use of construction robots in hazardous environments. In Section 4, the opportunities and challenges faced by construction robots operating in polluted and hazardous settings are comprehensively scrutinized. Section 5 focuses on the deployment of construction robots in demanding natural environments and provides an overview of current research in this area. Section 6 discusses practical implications, potential applications, and future prospects. Finally, Section 7 presents the concluding remarks for this article.
2. Approach to research and scientometric analysis
This article focuses on the use of construction robots in extreme environments and aims to establish a comprehensive framework for analyzing the relevant literature. By conducting a systematic retrieval and screening of relevant studies, the article provides an overview of the current research trends in construction robotics technology in extreme environments, and offers insights into the future development of pivotal technologies for construction robots operating in extreme environments.
2.1. Representative cases and key technologies
Robots play a crucial role in the architecture, engineering, and construction (AEC) sector, where they are widely used in various applications that involve multidisciplinary integration methods, digital tools, and innovative materials [28]. The success of construction robots relies heavily on the fundamental technologies of mechanism design, perception, planning, and control.
Mechanism design involves the creation of a detailed structural blueprint that aligns with the overall conceptual framework of the robot [29]. For example, Naclerio et al. [22] achieved control over lift and resistance in an underground robot by using an asymmetric structural design for the tip extension, allowing for navigable underground excavation (Fig. 2(a)). During the design phase, abstract operational principles are translated into specific components or parts. This includes specifying the material, shape, size, processing technology, strength, stiffness, and other relevant attributes of the structural elements, while also addressing precision and inter-component relationships [30], [31].
Perception plays a crucial role in a robot’s interaction with its environment, with visual perception being an important component of the robotic system. In complex scenarios, visual sensors such as cameras, light detection and ranging (LiDAR), and millimeter-wave radar are used to capture real spatial information about the surroundings. Through preprocessing, feature mapping, and analysis, a comprehensive representation of the external environment is obtained, providing the machine with rich 2D and 3D information. For instance, Johns et al. [32] used LiDAR to perceive highly heterogeneous in-situ stones and achieved online detection and segmentation of free-form stone instances depicted in Fig. 2(b). Furthermore, with the help of endpoint localization sensors, the in-situ construction robot AT40GW (Altec, USA) achieved precise 3D pose control, demonstrating mean pose repeatability error and mean pose repeatability of 13.2 and 6.9 mm, respectively (shown in Fig. 2(c) [33]). This comprehensive and high-quality environmental information enables the identification, tracking, and understanding of the characteristics and attributes of the surrounding environment, forming the basis for subsequent decision-making and operations.
In the dynamic and uncertain environment of construction, robots need to engage in motion planning. This involves determining a collision-free trajectory from the initial to the target pose point within the pose space, while considering various constraints such as environmental factors, time limitations, and the inherent dynamics constraints of the robot itself [14]. Additionally, the ability to plan job trajectories and allocate tasks is crucial for the efficient, stable, and safe completion of complex tasks in construction operations. For example, in the planning framework of robotic dry stone construction, Johns et al. [32] utilized LiDAR for sensing in-situ stones and generating a 3D point cloud image. They achieved stone grabbing and stacking based on a geometric planner shown in Fig. 2(b). Construction robot planning technology empowers robots to efficiently perform construction operations in complex environments, avoiding collisions and conflicts.
The primary goal of robot control is to facilitate the execution of predefined tasks and functions. One commonly used control strategy, known as proportional-integral-derivative (PID) control, provides stability and robustness. However, PID control has the drawback of complex parameter adjustment. In the case of time-varying and nonlinear systems, further optimization is required for PID-based control. The control process includes trajectory control, force control, attitude control, and more, all aimed at ensuring that the robot accurately and efficiently performs tasks according to planning requirements. The optimization of robot control models is a challenge in the construction process. To address this challenge, Zhang et al. [34] proposed a multi-agent control framework based on spatial collision perception and real-time autonomous task allocation, enabling minimally supervised multi-robot control with system robustness as exemplified in Fig. 2(d). Incorporating thoughtful feedback is crucial for control stability, and control methods driven by data and mechanism hybridization have advantages in practical applications. Robust control strategies are particularly important in unstructured scenarios for successful construction operations.
2.2. Bibliometric analysis
This study applies the preferred reporting items for systematic reviews and meta-analyses (PRISMA) method for literature review [35]. PRISMA includes a 27-item checklist and a flow diagram that clearly depict the process and fundamental principles of identifying, selecting, including, and excluding literature to enhance the accuracy of systematic review and meta-analysis reports. The PRISMA workflow diagram used in this investigation is from Refs. [36], [37], which is also shown in Fig. S1 in Appendix A.
To ensure the scientific integrity of the data sources [38], this research literature was systematically compiled from Web of Science, Scopus, and Google Scholar. The compilation of keywords related to extreme environments and construction robots involved merging the two sets using an ampersand (&). For instance, keywords associated with extreme environments included terms such as ((extreme environment) OR (dangerous operating environment) OR (polluted environment) OR (hazardous environment) OR (harsh environment)), while keywords related to construction robots included ((construction robotics) OR (autonomous construction machine) OR (unmanned construction machinery) OR (engineering robot)).
During the identification phase of this article, abstracts are examined based on the technical criteria of PRISMA. Within the PRISMA terminology, the full texts that are retrieved are referred to as reports and their eligibility is assessed. Irrelevant papers and review articles were rigorously excluded, resulting in a final database of 112 papers for further analysis.
The literature categories in this article have been systematically organized and are presented in Table 1. Robotics plays a crucial role in the field of construction, especially in challenging environments. Expertise in engineering, automation control systems, computer science, and physics is essential for various aspects such as perception, planning, control, and mechanism design. Strong engineering knowledge provides a solid foundation for effective planning, while automation control systems enable construction machinery and robotics to perform tasks with precision. In the domain of computer science, deep learning significantly enhances the intelligence of autonomous systems. Mechanism design involves the incorporation of physical and dynamic models. The subsequent sections of this article will explore these technical details, organized according to scenario divisions.
2.3. Keyword co-concurrence
Furthermore, keyword co-concurrence analysis is conducted to gain insights into the research themes and the interconnectedness of research endeavors. Keywords serve as concise representations of research topics and assist in analyzing the knowledge domain within the literature. Visualizing the network of keywords helps identify research hotspots and trends in the field of construction robots operating in extreme environments. Cluster topic analysis using the log-likelihood rate (LLR) algorithm in CiteSpace allows for the extraction of multiple clusters of literature, revealing emerging research directions within this domain. The keyword co-occurrence network is depicted in Fig. 3.
This article analyzes the classification of literature and identifies the top two keywords with the highest weight in each cluster. Then a chord diagram that illustrates the co-occurrence of these keywords is created, making it easier to identify prominent research areas (Fig. 3). Noteworthy keywords in this domain include engineering, construction technology, and robotics, highlighting their significance. Computer science also plays a crucial role in achieving environmental awareness, which is one of the analytical dimensions supported in this article. From the cluster analysis results, design, control, and path planning are identified as significant research dimensions. Additionally, the clustering results reflect various application scenarios, such as space robots, field robots, and decontamination, providing insights into the diverse challenges that construction robots face.
3. Construction robots in dangerous work environments
The widespread implementation of construction robots has the potential to reduce the exposure of construction workers to dangerous work environments. A comprehensive analysis across four key dimensions was conducted: blasting operations, demolition operations, post-disaster operations, and confined space operations, highlighting the crucial role of construction robots in enhancing safety in these domains.
3.1. Blasting work
Numerous tasks in the construction industry require blasting operations to be carried out, such as the excavation of tunnels and pipe networks, the crushing of large rocks during road paving, and the excavation for water conservancy projects. In the underground tunnel construction environment, Bonchis et al. [39] used a 3D laser rangefinder integrated into the robot to measure and create a tunnel model. They utilized the iterative closest point algorithm to calculate the position of the borehole relative to the robot, enabling the matching of blast holes and control of the robotic arm. Zhou et al. [40] proposed a hybrid compensation methodology to address position errors in rock drilling robots. The experimental results demonstrated that the average error of the mechanical structure control was 30.5 mm, with a standard deviation of 17.4 mm. This approach effectively mitigates non-geometric errors caused by extreme environmental factors like temperature variations, as depicted in Fig. 4 [40]. Among the various methods employed, the radial basis function network plays a significant role due to its three-layer network structure. By utilizing the input data x from essential sensors and the kinematic model parameters of the rock drilling robot, the predicted three-dimensional coordinates fi(x) can be calculated using Eq. (1). The basis function of the hidden layer and the connection weight of the output layer are denoted as ω and ϕ, respectively. The radial basis kernel function is computed using the Gaussian radial basis function (ϕk()) as specified in formula (2):
here ck represents the center of the basis function for the kth neuron node, σk denotes the expansion width of the radial basis function for the initial neuron in the hidden layer, i is the individual dimension, θ represents the rotation angle of two adjacent joints around the common axis, m is the number of dimensions of the input data, T represents matrix transpose, and wik is the ikth output layer connection weight.
Blasting construction is a hazardous operation that requires high control accuracy. Montorfano and Ruberto [41] implemented collaborative control of robotic drilling and Giti blasting through teleoperation, effectively carrying out construction tasks related to cross tunnels and safety niches. Drilling and blasting operations are part of the blasting activities. Nguyen and Nguyen [42] designed a blasting hole drilling robot arm equipped with a self-propelled hydraulic system with seven degrees of freedom. They developed a control algorithm based on kinematic principles to achieve automated blast hole drilling. The precision of blast hole drilling significantly impacts operational quality, as incorrect hole positions can introduce potential hazards. Additionally, the substantial weight and large span of the robot manipulator make precise control a challenging task. To improve the accuracy of blast hole drilling, it becomes crucial to address and compensate for control errors. Xie et al. [43] designed a cantilever blast hole drilling robotic arm and implemented a general regression neural network (GRNN) method to predict and correct directional errors, thereby achieving precise control of the robotic arm. The simulation results demonstrate that the maximum position error is less than 11 mm, and the range of deviation meets the requirements of engineering.
3.2. Demolition work
During demolition operations, there is a continuous cascade of construction debris and a significant amount of dust, making it a hazardous task. Demolition machinery utilizes hydraulic impactors for crushing and dismantling operations within the work area [44]. Additionally, there is a potential risk of the entire building structure collapsing. To ensure the safety of personnel involved in demolition activities and improve operational efficiency, remote operation is commonly utilized for most demolition robots. An example of such a remote-controlled demolition robot is the one developed by the German company Brokk, as shown in Fig. 5(a) [45], [46], [47]. Common actuators used in demolition robots include crushers [48], drum cutters [49], and surface grinders [50].
Deep reinforcement learning offers a stable approach for collision-free path planning in narrow-pipe redundant robots [51]. This involves an agent updating its state based on observed interactions with the environment and maximizing a reward function. In deep reinforcement learning, the agent updates its own state a based on the reward function r by observing s during the interaction with the environment. The objective of the agent is to maximize the reward function (Q*), which is defined by Eq. (3) [52]:
where π = P(a|s) is defined as behavior policy, P means that the state of the world changes probabilistically, t is time-step, and γ is discounted factor. $\mathbb{E}[]$ means the mathematical expectation. During the learning process, Q-learning can be applied to the sample (s, a, r, s′) ∼ U(D) to update, and the loss function (Li()) is defined as Eq. (4).
$L_{i}\left(\theta_{i}\right)=\mathbb{E}_{\left(s, a, r, s^{\prime}\right)!\sim \mathrm{U}(D)}\left[\left(r+\gamma \max _{a^{\prime}}\left(s^{\prime}, a^{\prime} ; \theta_{i}^{-}\right)-Q\left(s, a ; \theta_{i}\right)\right)^{2}\right]$
where θi and $\theta_{i}^{-}$ are the network parameters and target network parameters at the ith iteration, respectively [52]. Lee and Brell-Cokcan [46] proposed a control strategy optimization method based on reinforcement learning, enhancing the teleoperation construction efficiency of the Brokk 170 (Brokk, Sweden) construction robot, and q1,..., q5 are the joint space coordinates, as shown in Fig. 5(b). They utilized multilayer perceptron (MLP) to address the challenges of regression approximation posed by nonlinear models. Lee and Brell-Cokcan [47] further applied an MLP-based data-driven approach to implement motion trajectory prediction for the Brokk 170 construction robot, leading to improved construction efficiency and safety, and $u_{1}^{t}$ means the control input matrix of joint q1 at time t, as illustrated in Fig. 5(c).
Regarding morphological design, Li et al. [44] devised a four-degree-of-freedom boom for a demolition robot in order to improve its vibration characteristics. Corucci and Ruffaldi [53] addressed the perception challenges of demolition operations by developing a feedback-based scene segmentation method. Similarly, Zong et al. [54] tackled the challenges of target loss in demolition operations by implementing a 3D tracking algorithm. Huang et al. [48] achieved autonomous positioning of the end of the demolition robot’s manipulator arm using a laser-guided artificial neural network (ANN) global mapping model.
While current research primarily focuses on the structural design and perception capabilities of demolition robots, collision detection and dynamic control within the demolition operational environment receive relatively less attention. Residual vibrations during operations prove to be a critical challenge. Ye et al. [55] proposed a control methodology based on the exponential reaching law to effectively manage residual vibrations. Mu et al. [49] developed an enhanced directional bounding box method for real-time collision detection and implemented a composite control approach to maintain precise tracking accuracy within a 0.04 m threshold. Improvements in scene understanding and construction trajectory planning are crucial to enhance the autonomy and decision-making capabilities of construction robots in demolition construction planning.
3.3. Post-disaster work
Robots have the ability to manipulate heavy objects and navigate through complex terrains which makes them extremely valuable in post-disaster scenarios. In situations like earthquakes, floods, or mudslides, post-disaster tasks often involve working in unstructured construction environments, such as site cleanup, material transportation, dredging, and site restoration. Robots that are designed for post-disaster operations need to be able to adjust their posture flexibly and maintain stable performance. This is depicted in Fig. 6(a) [56].
To enhance the adaptability of robots in complex post-disaster environments, Kamezaki et al. [57] developed a remote-controlled robot with a four-arm, four-track configuration that is shaped like an octopus. This octopus-shaped robot effectively avoids overturning and performs exceptionally well on rugged and unstable terrain after disasters, as shown in Fig. 6(b). Yuta [58] created a heavy-duty transport robot with crawler tracks that is specifically designed for excavation tasks in the aftermath of floods.
To improve the perception capabilities of construction robots in post-disaster environments, Ootsubo et al. [59], [60] developed an augmented reality (AR)-based remote operating system for engineering robots. This system overlays 3D geometric information onto video images to enhance environmental perception. Nubert et al. [61] proposed an effective sensor fusion technique that integrates data from an inertial measurement unit (IMU) and LiDAR to provide reliable and accurate pose estimation. Fukui et al. [62], [63] and Yasuda et al. [64] achieved autonomous path planning for construction robots in hazardous post-disaster environments using a modular track structure. They also incorporated probabilistic roadmaps to dynamically adjust trajectories when encountering moving obstacles, thereby improving path planning in the presence of such obstacles.
Post-disaster environments are often characterized by intense vibrations and severe effects, which present significant challenges to construction robot control technology. In order to address the slipping problem that unmanned bulldozers face during dredging operations following floods, He et al. [65] proposed an anti-skid control technique that utilizes the active disturbance rejection control algorithm. Minamoto et al. [66] employed a force feedback-based remote control method to enhance the efficiency of damage recovery activities. However, existing research has not extensively focused on the design of robot bodies specifically for post-disaster operations, which often requires modifications to existing construction robots. Therefore, further development is needed to enhance the flexibility of robots to meet the demands of post-disaster operations.
3.4. Confined space work
Confined spaces within the construction sector include various environments such as infrastructure facility pipe networks, urban underground tunnels, sewage treatment facilities, and tightly sealed tunnels. These spaces often contain toxic and harmful gases, posing significant risks to workers. Therefore, the use of robots in construction tasks is necessary to ensure safety and efficiency [67].
In order to carry out tasks such as facility monitoring, maintenance, and construction in restricted space environments, specially designed robots with high maneuverability and flexibility are crucial. There are different types of robots commonly used in confined spaces, including peristaltic types [24], [68], [69], wheeled type [13], [70], [71], crawler type [72], and bionic type [23].
The complex layout of underground pipe networks, which often have inclinations, presents challenges for robot control stability. To address this issue, Mateos et al. [71] utilized proportional control technology in a six-wheeled legged robot, enabling it to perform pipeline repair tasks under intra-pipe pressure. Liu et al. [72] optimized the PID control of a crawler robot using swarm intelligence algorithms, resulting in improved step response performance.
Peristaltic robots, consisting of multiple structural units, are well-suited for operations in complex pipe network facilities. Tracked and wheeled robots possess the capability to navigate swiftly through pipelines. Bionic robots, known for their unique morphological designs, offer enhanced flexibility. For example, Ishikawa et al. [24] designed a peristaltic robot composed of seven structural units, enabling agile movement within narrow pipes. Miro et al. [13] employed Mecanum wheels to facilitate comprehensive motion within pipelines, as shown in Fig. 7(a). Qin et al. [23] used bionic principles to design a variable curvature trunk robot, improving the operational capabilities of the end effector in confined spaces when inspecting or repairing damaged areas.One efficient operational mode for pipe network robots is achieving active adhesion through the force between the driving wheel and the pipe wall, as illustrated in Fig. 7(b) [73].
Given that cameras alone may not provide sufficient environmental information in confined spaces due to obstacles, auxiliary sensing technologies based on sonar and ultrasonic radar are commonly employed. Hua et al. [74] utilized a real-time collision detection feedback gradient optimization algorithm for obstacle avoidance planning, preventing the working arm’s collision with the pipeline. Confined spaces often feature unevenly distributed obstacles, and collision-free path planning for narrow-pipe redundant robots can be achieved through reinforcement learning [49]. Zheng et al. [75] employed feedback control to enable rolling and yaw movements of each joint of the pipeline robot, allowing flexible obstacle avoidance in T-shaped pipelines. Hybrid wheel-legged robots can achieve obstacle avoidance based on a fuzzy logic controller.
4. Construction robots in polluted and harmful environment
Pollution and hazardous environments pose serious risks to workers’ health. Deploying construction robots can significantly reduce workers’ exposure to these dangers, protecting their lives and well-being. Although there is a significant demand for construction robots in polluted and hazardous environments, practical applications still face several key challenges.
4.1. Dust polluted environment
Environments such as mine tunnels and chemical plants often pose safety risks for construction workers due to toxic gases, harmful dust, and reduced visibility. Accurate environmental perception is crucial for working in such conditions to prevent hazardous accidents that can result in human casualties and property losses. Therefore, addressing the issue of environmental perception in high-dust environments is essential for construction robots. The development of computer vision has provided solutions for robot perception in high-dust and low-light environments. A notable research area within computer vision is image enhancement technologies designed for visually degraded environments. By utilizing the statistical principle of dark channel prior, image de-fogging can be achieved. This technique shows great potential for enhancing visual clarity in dusty environments [76], [77], and the detailed introduction can be found in the Appendix A.
LiDAR technology is widely used in robotic systems to obtain precise distance perception information. However, adverse environmental conditions such as fog and rain can have an impact on the transmission loss T(R) of LiDAR, as illustrated in Fig. 8(a) [78]. The measurement range of the LiDAR, denoted as R, is calculated using Eq. (5), taking into account the time-of-flight (ToF) principle. Additionally, the transmission loss can be determined using Eq. (6) [78]:
where c is the speed of light, tr is the transmission medium, n and α(tr) stand for the refractive index and extinction coefficient of the medium through which propagation occurs, respectively, and Δt denotes the time interval between the emission and reception of the laser.
Environmental factors can have negative effects on the performance of construction robots. Incomplete and sparse point clouds, as well as LiDAR noise, can hinder the perception capabilities of these robots. In order to address these challenges, Emter et al. [79] enhanced the density of point clouds by utilizing data fusion techniques with multiple LiDARs located at different positions. This led to improved recognition accuracy for construction robots. Lin et al. [80] tackled the issues of dust and occlusion by obtaining 3D point cloud data through 2D LiDAR rotation scanning. Additionally, Ouyang et al. [81] achieved perception in underground environments by stitching LiDAR point clouds and reorienting normals. Deep learning approaches, like VoxelNet [82], have shown effectiveness in extracting local features from point clouds, thereby enhancing point cloud registration results. Representative examples of mine roadways and narrow tunnels are shown in Fig. 8(b) [83] and Fig. 8(c) [14], respectively.
4.2. Solid waste polluted environment
The generation of solid waste from human activities poses contamination risks as it can consist of various materials, ranging from household waste like batteries and paints to industrial waste such as gypsum and soil. Kruse and Wilson [84] utilized remotely controlled excavation robots for hazardous waste burial operations, contributing to environmental restoration efforts. The robot systems for decontamination in hostile environments (ROBDEKON) project has made significant contributions to the design of robot ontologies for solid waste pollution treatment. As part of this project, crawler excavators have been developed with the capability to navigate landfill sites, retrieve objects, autonomously remove contaminated soil layers, and efficiently transfer materials. This enhances the robot’s performance in complex and muddy landfill areas, making it more versatile and stable [85]. To prevent secondary pollution during solid waste handling, Emter et al. [79] employed an approximate probabilistic algorithm for motion planning of excavation manipulators with a high degree of freedom.
Obstacles frequently occur in excavation sites, presenting as large, rigid objects or accumulated debris. These obstructions can impede the progress of excavation operations. In order to address this challenge, Zhang et al. [1] have developed an autonomous excavator system (AES), which is specifically designed for waste treatment and recycling operations. They utilized Mask R-CNN algorithm to detect hindrances within the site, with a particular focus on blocking objects, as shown in Fig. 9. Robots equipped with accurate environmental maps and positioning capabilities [79] can autonomously carry out practical tasks such as removing wall materials or decontaminating factory components. ROBDEKON tackles the issue of precise geometric measurement of objects to ensure robots’ efficient functioning [85]. For robust and accurate environmental perception of autonomous engineering machinery, Heide and Petereit [86] proposed an enhanced combination method that utilizes custom validation techniques and machine learning methods.
4.3. Nuclear radiation polluted environment
The incidents at Chernobyl and Fukushima have highlighted the necessity of entering environments contaminated by nuclear radiation for various operations. Robots, capable of performing complex tasks in high-radiation settings, can effectively replace humans in tasks such as facility repair, maintenance, decommissioning, and site cleanup, as depicted in Fig. 10 [16]. However, it is important to note that robots are vulnerable to damage from nuclear radiation, necessitating special optimization of their physical structures. The charge-coupled device (CCD) camera CY-RC51KD (Panasonic, Japan) exhibited a change in image quality from clear to bluish when the radiation level exceeded 140.0 Gy and was rendered inoperable after reaching 169.0 Gy. The laser scanner (URG-04LN, Hokuyo, Japan) ceased functioning after accumulating a total dose of 124.2 Gy. However, the laser scanner models UXM-30LN (Hokuyo, Japan) and Eco-scan FX8 (Nippon Signal, Japan), as well as the CCD camera model Axis 212, were able to withstand a total dose of 200 Gy [16].
Nuclear environments often involve complex tasks within dense concrete structures and cluttered surroundings. In order to meet the demands of flexible operations in such settings, Bird et al. [87] have designed the Vega robot, which integrates a five-degree-of-freedom manipulator. West et al. [67] have developed a platform that combines a mobile base with a compact robotic arm, enabling remote operation and a wide range of tasks. Crawler robots equipped with multifunctional manipulators are also well-suited for handling complex tasks in environments exposed to nuclear radiation [85], [88].
Effective path planning is crucial in enabling robots to move quickly and minimize their exposure to nuclear radiation. Previous robots have encountered challenges in nuclear environments, such as debris entanglement or damage. Lai and Smith [89] have developed biomimetic path planning strategies to reduce radiation exposure. Particle swarm optimization (PSO) and genetic algorithms have both demonstrated their ability to identify smoother paths with lower radiation doses compared to grid-based algorithms. Chao et al. [90] have introduced the DL-RRT* algorithm, a sampling-based approach that determines paths with minimal radiation dose in dynamic environments. This algorithm leverages information from the previous planning phase, reducing the need for global re-planning and, consequently, decreasing the robot’s travel time in radiation environments. The robot operating system provides robust system support for autonomous navigation and sensor integration [91].
Nuclear radiation environments often require the execution of complex tasks with a high level of accuracy, necessitating the optimization of the robot control system to meet specific requirements. Bakari et al. [92] have achieved real-time control of two multi-functional arms for tasks such as gripping and cutting, using the PID control method. Components within nuclear facilities are characterized by elevated levels of residual radioactivity and temperature. Combined with their large size, mass, and complex shapes, precise trajectory tracking control is crucial while using remote-controlled manipulators for operations. In this context, Li et al. [93] have proposed a fuzzy logic system designed to enhance control performance for a five-degree-of-freedom robot manipulator used in nuclear reactor dismantlement. The aim is to improve the accuracy of trajectory tracking control during maintenance operations in nuclear radiation facilities. Luo et al. [94] have adopted a human-machine collaborative control strategy to achieve automatic welding with a tracking accuracy of ± 0.1 mm.
5. Construction robots in harsh natural environment
Human curiosity continues to drive our relentless pursuit of exploration, even in the face of the most inhospitable natural environments. Among these environments, the deep sea, underground domains, and outer space stand as prime examples, each presenting its own unique and formidable challenges.
5.1. Deep underground
The deep underground environment poses complex and challenging obstacles, including extreme earth pressure and the constant threat of geological collapse [95]. This creates significant operational difficulties for construction robots. In order to ensure the safety and efficiency of tunneling operations in such conditions, shield machines are crucial equipment for tunnel construction. Shield tunneling is widely recognized as a state-of-the-art method for constructing subway tunnels. Armaghani et al. [96] proposed an intelligent predictive model to evaluate the performance of tunnel boring machine (TBM), as shown in Fig. 11(a). Gong et al. [97] have proposed an improved variable gain position/speed control method that demonstrates superior grasping performance on both rigid and flexible targets. This method is well-suited for operations in deep and extreme environments characterized by complex geological conditions. The process of building shield machines involves gathering a wide range of multidimensional and diverse data as shown in Fig. 11(b) [15]. Accurately extracting and analyzing this data with reliable modeling techniques is of utmost importance.
In the planning of excavation tasks, Homma et al. [98] utilized tactile sensors to gather soil characteristics from deep areas, which are then used to calculate the optimal excavation parameters. On the other hand, Zhang et al. [99] proposed an intelligent decision-making approach that utilizes the PSO algorithm to determine the most suitable mining parameters. This decision-making method has proven its effectiveness and reliability in practical applications.
Intelligent optimization methods, such as the PSO algorithm [100], are effective for achieving global optimization in complex problems and offer high accuracy, and the detailed introduction to PSO can be found in the Appendix A. In order to navigate diverse working conditions in different geological sections, deep-earth robots need to develop reliable motion plans by integrating various sources of information, including robot arm position, force conditions, and geological data. Lee et al. [25] have developed a mole-like excavating robot, which includes an expandable drill bit mechanism. This innovative robot imitates the strong tooth clamping force of a mole and incorporates chisel teeth with a high level of flexibility, enabling it to excavate more resilient ground and create wider-diameter holes that allow for easier movement (Fig. 11(c)). Conventional deep excavation operations often rely on large machines with rigid and oversized components to overcome granular drag and granular force lift. However, Naclerio et al. [22] have successfully used a soft robot that can achieve high-speed penetration of the ground and underground excavation by controlling underground interaction forces. Operating in complex deep underground environments can lead to abnormal geological effects, presenting challenges for robot control.
5.2. Deep sea
Deep sea construction robots encounter numerous technical obstacles, with heightened water pressure being a major concern. The immense water pressure found in the ocean’s depths poses a significant threat to the robots’ structural integrity and materials [101]. To ensure their effectiveness and safety, the robots must be designed to withstand these extreme conditions, as shown in Fig. 12(a) [102]. Mechanism design plays a crucial role in optimizing the sealing performance of deep sea construction robots under high water pressure, allowing them to perform maintenance tasks on underwater structures [103]. Multi-joint peristaltic robots [104], [105] are commonly used for deep sea geological exploration and small-scale excavation projects. Nagai et al. [105] proposed a robot that utilizes creeping crawling to excavate the seabed. This robot consists of three components—an excavation unit, a propulsion unit, and an extraction unit—and can handle water pressure up to 5 MPa. However, designing mechanisms for such extreme environments is challenging, as shown in Fig. 12(b), because the engineering equipment can lose its structural integrity and sustain irreparable damage [106]. Tadami et al. [104] developed a robot specifically designed for seabed excavation in underwater exploration, featuring a propulsion unit and an excavation unit that enables efficient navigation within the seabed.
Apart from high water pressure, deep sea construction robots also face difficulties due to low lighting conditions and image distortion, which significantly affect perception. The deep sea experiences minimal natural light, resulting in extremely limited environmental illumination. As a result, sensors and vision systems are severely restricted, necessitating the use of specialized technologies like sonar and infrared sensors to compensate for these limitations. For example, Kim et al. [107], [108] equipped robots with sonar, gyroscopes, and depth sensors to gather complex environmental information for leveling construction of submarine gravel sites. Two-dimensional multi-beam imaging sonar is employed to visualize reflection intensity data. Additionally, Song and Choi [109] proposed a method that utilizes 2D multi-beam imaging sonar to reconstruct underwater building structures in three dimensions, providing high-resolution images in deep sea environments. The poor water quality in underwater environments can cause image distortion and blurring, further complicating visual recognition. To overcome this issue, AncutiT et al. [110] developed a method to enhance underwater images by calculating the local backscattered light of the image, which was successfully validated through practical testing.
Construction tasks in underwater environments may encounter challenges due to increased turbidity levels. This can hinder robots from accurately perceiving and navigating the construction scene, affecting construction quality and efficiency. Addressing these challenges requires comprehensive research and ongoing innovation to ensure the effective performance of deep sea construction robots in extreme underwater conditions [111].
The deep sea is known for its complex currents and unpredictable ocean conditions, which can disrupt operations. As a result, robots need to possess high stability and adaptability to maintain precise position and attitude. Hong et al. [112] utilized the pure-pursuit method to calculate the turning radius, enabling underwater construction robots to follow a designated path and counteract yaw disturbances in the deep sea. Conventional control methods that rely on numerical techniques, kinematic approaches, and geometric strategies may not always be sufficient for dynamic and changing conditions [113].
To enhance the propulsion efficiency and agility of robots in underwater disturbance environments and facilitate reliable maintenance of underwater infrastructure, Gorma et al. [114] proposed a bionic control method based on the central pattern generator. Xia et al. [115] proposed a layered intuitive control method based on virtual reality and a tactile simulator, improving operator perception and the teleoperation performance of undersea robots. Kim et al. [116] used active control technology to maintain optimal torque and speed in deep sea trenching operations under a water pressure of 20 MPa. Model-based meta reinforcement learning provides adaptability to environmental changes and disturbances [117]. Moon et al. [118] presented a model-based reinforcement learning approach and a manipulator control method suitable for underwater robots, validated through simulations.
5.3. Deep space
The deep space environment includes a combination of extreme characteristics, necessitating the integration of various fields such as astrophysics, materials science, engineering, and computer science. Robots play a crucial role in supporting human activities related to the construction and maintenance of extraterrestrial bases and exploration [119], [120].
In the context of extreme extraterrestrial environments [121], [122], [123], challenges arise due to factors like low light [124] and space stray light. To tackle these challenges, Sun et al. [125] utilized a pose estimation algorithm based on Yolov3 and the vision-flow fusion network (VFFNet) to accurately determine the position of the load plate under complex lighting conditions. Infrared sensors are employed to overcome the limitations of visible light perception. Zhou et al. [126] implemented a sensing system with binocular cameras and infrared radiation sensors in their extraterrestrial construction robot.
Extreme environments in deep space, such as the Moon and Mars, are highly uncertain and pose unique challenges due to the limited understanding of these environments by humans. Traditional motion planning methods, such as the roadmap method [127] and rapidly-exploring random tree (RRT) [128], have limitations in dealing with high uncertainty and lack adaptability. To address planning problems in multi-constraint environments, Yue et al. [129] proposed a hierarchical path planning approach specifically designed for navigating complex spatial environments.
In the microgravity environment of space, vibrations pose a significant threat to large structures. Manipulators, being lightweight and flexible, are particularly susceptible to vibrations during space operations [9]. This can greatly compromise positioning precision, making vibration control a critical issue [130]. To address this challenge, various methods of active compliance control, such as traditional “force-displacement” hybrid control and impedance control, are commonly used in the construction of large space structures. Boning and Dubowsky [131] apply linear optimal control theory to manage space structure construction robots, with the objective of reducing structural vibrations. Their approach enables robots to efficiently navigate and assemble sizable, flexible space structures while minimizing vibrations. Additionally, the introduction of an optimal linear quadratic regulator method helps mitigate vibrations in flexible manipulators [132]. Numerical simulations have validated the effectiveness of this method in ensuring closed-loop stability, even in the presence of unknown disturbances. Meanwhile, Sun et al. [125] propose a technique that combines vision guidance with variable parameter impedance control to effectively suppress oscillations during the construction process. Galardini et al. [133] employ compliance control and virtual depth control for the ETS-VII robot arm in lunar structure assembly tasks, ultimately achieving improved control precision. Furthermore, studies have identified multi-functional robots equipped with excavation robotic arms and bulldozer blades as particularly suitable for lunar site work [133], [134], [135]. Some typical construction robots working in deep space are shown in Fig. 13 [120], [136], [137].
6. Discussion
6.1. Challenges applied to practice
6.1.1. Adaptation mechanism design
Extreme environments include a wide range of challenging work conditions, such as hazardous locations like blast sites, demolition areas, disaster recovery zones, and confined spaces. These environments often present potential dangers that can jeopardize the safety of human construction workers. In contrast, robots exhibit remarkable proficiency in performing tasks in contaminated and hazardous settings. They are adept at addressing challenges such as dust pollution, solid waste contamination, and nuclear hazards. Additionally, robots demonstrate versatility in adapting to harsh natural terrains, including deep subterranean surveying and detection, deep sea maintenance and scientific research, and even deep space exploration missions. By utilizing specialized mechanisms, robots can effectively operate in environments that are unsuitable for human presence, ensuring both safety and efficiency in construction processes [4]. These extreme environments have unique physical characteristics that require robots to perform a variety of tasks, such as lifting heavy objects and navigating obstacles, thereby introducing complex and dynamic challenges in construction scenarios [1], [2].
6.1.2. Accurate perception of changing scenes
Accurate environmental perception plays a crucial role in protecting individuals and property during construction activities. Dust or haze can obscure targets and limit visibility at construction sites, increasing the risk of hazardous incidents. The perception abilities of deep sea construction robots are also challenged by low-light conditions and image distortion. In the deep sea domain, limited light penetration creates an environment with minimal illumination [5]. As a result, traditional vision sensors and systems perform poorly under these circumstances. To overcome these challenges, robots need to utilize specialized sensing technologies such as sonar and infrared sensors to compensate for the deficiencies in visual perception. Sonar technology enables the detection of targets and measurement of distances by utilizing sound wave propagation in water, providing reliable sensing capabilities in deep sea environments. Infrared sensors detect thermal radiation and are unaffected by lighting conditions, allowing them to maintain efficient sensing performance even in low-light or completely dark surroundings [2]. In extreme extraterrestrial environments, robots face similar issues with poor lighting conditions and spatial stray light interference [6]. These challenges significantly hinder a robot’s perception accuracy on planetary surfaces or in outer space.
6.1.3. Intelligent planning for unstructured sites
The intricacies and challenges surrounding intelligent decision-making in unstructured environments are many. In the context of in-situ construction in extreme extraterrestrial domains like the Moon and Mars, the high level of environmental uncertainty poses a significant obstacle for traditional motion planning methods, such as roadmap methods [127] and RRTs [128]. These conventional approaches have limitations when it comes to dealing with unfamiliar and dynamic environmental conditions, which make decision-making in these extreme settings exceptionally complex. Deep reinforcement learning learns through the machine’s autonomous interaction with the environment, which allows the machine to progressively gain experience and refine decision-making strategies to deal with unpredictable construction scenarios. This methodology utilizes deep neural networks and reinforcement learning algorithms to enable robots to continuously improve their intelligence and adapt to various operational environments. Intelligent decision-making can effectively navigate the complexities, uncertainties, and unpredictabilities present in construction scenarios, thereby ensuring the safety and efficiency of construction tasks. The ongoing development and improvement of these methods will lay a strong foundation and enhance the future of intelligent decision-making for robots operating in deep space and other extreme situations.
6.1.4. Optimization of control models
Optimizing the control model of robots is vital in extreme environments to ensure the desired control results. Robotic arms play a critical role in construction in extreme contexts. However, in microgravity conditions of extraterrestrial environments, the issue of robotic arm vibrations becomes particularly significant. Flexible manipulators, being lightweight, are prone to inducing substantial vibrations during spatial operations [9], potentially affecting precise positioning. Optimizing the control model focuses on suppressing vibrations to enhance system performance [125].
Environmental disturbances such as vibrations and severe effects are common in post-disaster scenarios, posing significant challenges to the control technology of construction robots. In nuclear facilities, components often have high residual radioactivity levels and operate at high temperatures. Remote-controlled manipulators are used for handling nuclear waste, which is bulky, heavy, and has complex shapes. Precise trajectory tracking control is crucial to ensure safety and accuracy in these operations. Accuracy is also vital in hazardous working environments like blast hole drilling, as incorrect hole placement can lead to potential hazards. Controlling the movements of heavy manipulators with large spans is a formidable task. To address possible interference in extreme settings, specialized design and optimization of the robot’s control model and algorithms are necessary. Continual improvement and refinement of these control technologies will establish a strong technical foundation for deploying robots in diverse extreme environments, enabling safer, more efficient, and more precise construction and operational activities.
6.1.5. Economic and operational implications
Construction tasks in the current AEC industry are often considered to be dirty, dangerous, and dull [135]. However, due to labor shortages, the cost of employment in the construction industry is steadily increasing [33], [135]. Construction in extreme environments poses even higher labor costs and safety risks compared to traditional construction processes [138]. In order to address these challenges, unmanned construction machinery and robots present new opportunities. By employing teleoperation, operators can work remotely and avoid dangerous or polluted environments [52], [59]. Additionally, the incorporation of autonomous decision-making capabilities into unmanned construction equipment minimizes the need for human intervention throughout the construction process [1]. As a result, the utilization of robotic technology in the AEC industry has the potential to reduce labor costs and enhance construction efficiency.
6.1.6. Safety features and ethical framework
The construction safety of extreme construction robots is challenged by the complexity of human-machine interaction [139]. When designing, researching, testing, and deploying extreme environment construction robots, it is essential to establish an ethical framework. To address this, safety assessments need to be conducted on robot stability, reliability, and other aspects. The fusion of multi-source heterogeneous sensors can enhance the safety of extreme construction robots by overcoming the limitations of individual sensor performance [140], [141]. Furthermore, the design of redundant systems can ensure that the robot can continue working even if some components fail [57]. Prioritizing personnel safety, robot operations can be immediately halted in the event of severe failures or hazards, or remotely taken over. Research efforts should fully consider safety issues in extreme environments and propose safety standards and protective mechanisms. This is crucial to prevent endangering human lives during robot operations in hazardous workplaces, polluted areas, and adverse natural environments [45].
6.2. Prospects and future outlook
6.2.1. Nature-inspired bionic design
Robots have significant potential for use in extreme environments such as outer space, undersea, and disaster-affected areas. They have already shown the ability to perform various tasks in challenging settings, easily navigating through aerial, aquatic, and land-based landscapes. Natural evolution has given subterranean creatures unique biological structures that reduce resistance and enhance propulsion. The Scincus scincus and Chionactis occipitalis are examples of cave-dwelling creatures that provide valuable insights for the design and modeling of robots [22].
The design of nature-inspired robots requires collaboration between experts in mechanical engineering, materials science, and computer-aided design. Interdisciplinary workshops and collaborative simulations can facilitate knowledge exchange and joint problem-solving. Peristaltic crawling robots, inspired by the locomotion of earthworms, achieve stable movement in narrow, confined spaces [24]. Elephant trunk robots, modeled after elephants, can navigate through confined and obstacle-ridden environments to perform complex tasks [23]. Drilling robots, designed like moles, play important roles in extreme conditions such as space and extraterrestrial construction [25]. Soft robots, with their adaptive shape-changing abilities, increase operational efficiency and precision by conforming to physical contact [12]. Legged robots, which use limbs as supports and mobile appendages, show superior resilience in unfamiliar environments compared to conventional wheeled robots [4]. In unstructured terrains like mud, gravel, snow, vegetation, and desert, legged robots demonstrate superior traversal capabilities [2].
Research into the design and modeling of robots for subterranean and unstructured environments not only expands the range of robot applications in challenging conditions but also provides valuable lessons for the development and improvement of future robotic technology. Continued exploration in this area will contribute to the integration of robots across diverse demanding environments and advance intelligent systems through automation.
6.2.2. Multimodal data fusion perception
The integration of multimodal data is of great importance for enhancing the cognitive abilities of autonomous systems. A robot’s perceptual abilities span various critical areas, including self-kinematics estimation, contact modeling, and mapping and understanding of its surroundings [11]. While computer vision plays a crucial role in enabling robots to gather information about their environment, robots face unique challenges in seamlessly integrating different sensory modalities, such as vision and touch, unlike humans [142]. By combining multimodal perception sensors like LiDAR and depth cameras, significant advancements can be made in areas such as material and texture classification, as well as acquiring semantic information [1]. Taking inspiration from the human brain, neuromorphic general place recognition system (NeuroGPR) is an emerging technology that imitates neuronal dynamics to achieve location recognition through multi-modal sensing within a unified spatial and temporal framework [143]. Nature provides us with various organisms capable of recognizing individuals across different sensory modalities, offering valuable insights for robots in fusing multimodal perception data [144]. Research in this field holds promise for supporting robot perception and decision-making across various application domains, driving the continuous development and refinement of robot technology.
6.2.3. Adaptive planning
In decision-making, humans rely on their experience and current environmental information to form intuitive, subjective decisions. This deep understanding of the physical world is crucial for construction robots to make effective plans and decisions. With accurate terrain data, a robot’s choice of foothold and trajectory becomes the key to task decision-making [2]. Existing research shows that footholds can be calculated more efficiently using heuristic methods [26] or convolutional neural networks [145]. Combining inverse reinforcement learning with imitation learning allows robots to learn strategies from human demonstrations [1]. This approach extracts valuable decision rules from human expertise, enhancing the intelligent decision-making of engineering machinery. By mimicking human decisions and actions, robots can better adapt to complex environments and improve their decision-making abilities.
Another promising research direction is exploring coordination mechanisms inspired by natural group behavior to achieve decision-making in robot swarms. Animals like birds and fish demonstrate remarkable collaboration and coordination, which can be applied to robot groups for more efficient task distribution and coordinated actions [27]. Studying natural group behavior offers new insights for intelligent decision-making and collaboration in robot swarms.
Adaptive decision-making, rooted in expert knowledge, plays a crucial role in advancing the intelligence of both individual construction machinery and robot clusters. These research areas hold the potential for significant innovations in robot decision-making and planning, ultimately improving robot performance in complex tasks and diverse environments.
6.2.4. Hybrid data-driven and mechanism-based control
The hybrid data-driven and mechanism-driven control approach offers innovative and viable solutions for the challenges faced in extreme environments. When dealing with control problems involving variable objects, an effective strategy involves integrating model priors and task data for perception and operation [146]. Data-driven methods, particularly reinforcement learning, have shown exceptional performance in handling complex dynamic situations while maintaining real-time control efficiency. These methods have made significant progress in controlling bipedal or quadruped robots, especially in simulated environments [16], [17]. Reinforcement learning enables the natural evolution of control strategies, allowing successful transfer of quadruped robot control from simulations to the real world [147]. Furthermore, by incorporating proprioceptive feedback into motion control, highly adaptive motion control of quadruped robots on complex terrains becomes achievable [148].
In addition to data-driven approaches, emulating the agile movements of animals in nature has proven effective in enhancing robot motion control performance. The attention-driven recurrent encoder integrates proprioceptive and external sensory inputs, enabling comprehensive end-to-end motion control. This method allows the robot to sense terrain characteristics before making contact, plan ahead, and adapt its gait, thereby enhancing its adaptability to different environments in the wild [2], [149]. Significant progress has been made in this research field, providing valuable inspiration and guidance for future robot control in complex and uncertain environments. By combining data-driven methods, reinforcement learning, and biologically inspired technologies, we anticipate further innovations and breakthroughs in addressing robot motion control challenges and establishing a solid foundation for improving robot performance across diverse tasks.
7. Conclusions
Extreme environments include hazardous workplaces, polluted and dangerous areas, and severe natural surroundings. Construction in these extreme contexts often involves risks that endanger the safety of human workers. Construction robots have emerged as a promising solution for exploring and working in these extreme domains. This article systematically examines construction robotics in diverse extreme environments, considering four key analytical dimensions: mechanism design, perception, planning, and control.
Our objective is to provide a thorough analysis of the technical challenges and knowledge gaps that surround the use of construction robots in extreme environments. Modeling and optimizing construction robots is a challenging task due to the uncertainty of the environment and the complexity of the data involved. However, research in robot kinematics modeling and biomimetic design has unearthed new possibilities for deploying robots in extreme conditions. Robots with specialized structural designs have demonstrated exceptional performance in these environments, with the ability to adapt their posture to suit different tasks and scenarios. Intelligent decision-making strategies are crucial for successfully handling complex and unpredictable construction scenarios, ensuring safety and efficiency. Further development and enhancement of these strategies will provide a strong foundation for future intelligent decision-making by robots, specifically in deep space and other extreme environments. Artificial intelligence techniques, particularly reinforcement learning, have proven resilient in controlling construction robots. Through a rigorous scientific analysis of existing research, we have identified potential avenues for the growth and expansion of this field. These pathways include ➀ designing robots inspired by nature, ➁ integrating multimodal data for perception, ➂ adapting decision-making strategies, and ➃ employing a hybrid approach that combines data-driven and mechanism-based control methodologies.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This research is supported in the Strategic Research and Consulting Project of the Chinese Academy of Engineering (2023-XZ-90 and 2023-JB-09-10), the National Key Research and Development Program of China (2021YFF0500301 and 2023YFB3711300), the National Natural Science Foundation of China (72171092 and 71821001), and the Natural Science Fund for Distinguished Young Scholars of Hubei Province (2021CFA091).
ZhangLJ, ZhaoJX, LongPX, WangLY, QianLF, LuFX, et al. An autonomous excavator system for material loading tasks. Sci Robot2021; 6(55): eabc3164.
[2]
MikiT, LeeJ, HwangboJ, WellhausenL, KoltunV, HutterM. Learning robust perceptive locomotion for quadrupedal robots in the wild. Sci Robot2022; 7 (62):eabk2822.
[3]
LuC, CaiC. Challenges and countermeasures for construction safety during the Sichuan-Tibet railway project. Engineering2019; 5(5):833-8.
[4]
ArmP, WaibelG, PreisigJ, TunaT, ZhouRY, BickelV, et al. Scientific exploration of challenging planetary analog environments with a team of legged robots. Sci Robot2023; 8(80):eade9548.
[5]
SmithKL, ShermanAD, McGillPR, HenthornRG, FerreiraJ, ConnollyTP, et al. Abyssal benthic rover, an autonomous vehicle for long-term monitoring of deep-ocean processes. Sci Robot2021; 6(60):eabl4925.
[6]
GaoY, ChienS. Review on space robotics: toward top-level science through space exploration. Sci Robot2017; 2(7):eaan5074.
[7]
LiuD, WangH, ZhongB, DingL. Servitization in construction and its transformation pathway: a value-adding perspective. Engineering2022;19:166-79.
[8]
ChenX. Research on combined construction technology for cross-subway tunnels in underground spaces. Engineering2018; 4(1):103-11.
[9]
YangG, ZhangLR, YuS, MengSC, WangQS, LiQJ. Influences of space perturbations on robotic assembly process of ultra-large structures. Nonlinear Dyn2023; 111(11):10025-48.
[10]
YouK, ZhouC, DingL. Deep learning technology for construction machinery and robotics. Autom Constr2023;150:104852.
[11]
ThuruthelTG, ShihB, LaschiC, TolleyMT. Soft robot perception using embedded soft sensors and recurrent neural networks. Sci Robot2019; 4(26): eaav1488.
[12]
ShihB, ShahD, LiJX, ThuruthelTG, ParkYL, IidaF, et al. Electronic skins and machine learning for intelligent soft robots. Sci Robot2020; 5(41):eaaz9239.
[13]
MiroJV, UlapaneN, ShiL, HuntD, BehrensM. Robotic pipeline wall thickness evaluation for dense nondestructive testing inspection. J Field Robot2018; 35 (8):1293-310.
[14]
LiuN, ChenK, DengE, YangWC, WangYW. Study on dust suppression performance of a new spray device during drilling and blasting construction in the metro tunnel. Tunn Undergr Space Technol2023;133:104975.
[15]
ZhouC, GaoY, ChenEJ, DingL, QinW. Deep learning technologies for shield tunneling: challenges and opportunities. Autom Constr2023;154:104982.
[16]
NagataniK, KiribayashiS, OkadaY, OtakeK, YoshidaK, TadokoroS, et al. Emergency response to the nuclear accident at the Fukushima Daiichi Nuclear Power Plants using mobile rescue robots. J Field Robot2013; 30 (1):44-63.
[17]
LiGZ, LiuSQ, WangLQ, ZhuR. Skin-inspired quadruple tactile sensors integrated on a robot hand enable object recognition. Sci Robot2020; 5(49): eabc8134.
[18]
MingboD, TaoM, JiajiaC, PanZ, HuaweiL. RRT-based motion planning algorithm for intelligent vehicle in complex environments. Robot2015; 37 (4):443-50.
[19]
YouK, ZhouC, DingL, ChenW, ZhangR, XuJ, et al. Earthwork digital twin for teleoperation of an automated bulldozer in edge dumping. J Field Robot2023; 40(8):1945-63.
[20]
PengXB, BersethG, van de Panne M. Terrain-adaptive locomotion skills using deep reinforcement learning. ACM Trans Graph2016; 35(4):1-12.
[21]
PengXB, BersethG, YinKK, Van De Panne M. DeepLoco: dynamic locomotion skills using hierarchical deep reinforcement learning. ACM Trans Graph2017; 36(4):1-13.
QinGD, WuHP, JiAH. Variable-curvature elephant trunk robot in nuclear industry. Fusion Eng Des2023;192:113642.
[24]
IshikawaR, TomitaT, YamadaY, NakamuraT.Development of a peristaltic crawling robot for long-distance complex line sewer pipe inspections. In:Proceedings of the IEEE International Conference on Advanced Intelligent Mechatronics (AIM); 2016 Jul 11-14; Banff, AB, Canada. Piscataway: IEEE; 2016. p. 413-8.
[25]
LeeJ, KimJ, MyungH. Design of forelimbs and digging mechanism of biomimetic mole robot for directional drilling. In:MajeedAPA, Mat-JizatJA, HassanMHA, TahaZ, ChoiHL, KimJ, editors. RITA2018. Lecturenotes in mechanical engineering. Singapore: Springer; 2019. p. 341-51.
BerlingerF, GauciM, NagpalR.Implicit coordination for 3D underwater collective behaviors in a fish-inspired robot swarm. Sci Robot 2021; 6(50): eabd8668.
[28]
YangGZ. Digital architecture and robotic construction. Sci Robot2017; 2(5): eaan3673.
PhelpsS, McBurneyP, ParsonsS. Evolutionary mechanism design: a review. Auton Agent Multi-Agent Syst2010; 21(2):237-64.
[32]
JohnsRL, WermelingerM, MascaroR, JudD, HurkxkensI, VaseyL, et al. A framework for robotic excavation and dry stone construction using on-site materials. Sci Robot2023; 8(84):eabp9758.
[33]
KeatingSJ, LelandJC, CaiL, OxmanN. Toward site-specific and self-sufficient robotic fabrication on architectural scales. Sci Robot2017; 2(5):eaam8986.
[34]
ZhangKT, ChermprayongP, XiaoF, TzoumanikasD, DamsT, KayS, et al. Aerial additive manufacturing with multiple autonomous robots. Nature2022; 609 (7928):709-17.
[35]
PageMJ, McKenzieJE, BossuytPM, BoutronI, HoffmannTC, MulrowCD, et al. The PRISMA2020 statement: an updated guideline for reporting systematic reviews. Int J Surg 2021;88:105906.
[36]
ZhaoYF, TaibN. Cloud-based building information modelling (Cloud-BIM): systematic literature review and bibliometric-qualitative analysis. Autom Constr2022;142:104468.
[37]
ChungSH, MoonS, KimJ, KimJ, LimS, ChiSK. Comparing natural language processing (NLP) applications in construction and computer science using preferred reporting items for systematic reviews (PRISMA). Autom Constr2023;154:105020.
[38]
PalA, HsiehSH. Deep-learning-based visual data analytics for smart construction management. Autom Constr2021;131:103892.
[39]
BonchisA, DuffE, RobertsJ, BosseM. Robotic explosive charging in mining and construction applications. IEEE Trans Autom Sci Eng2014; 11(1):245-50.
[40]
ZhouXY, BaiWY, HeJL, DaiJ, LiuP, ZhaoYM, et al. An enhanced positional error compensation method for rock drilling robots based on LightGBM and RBFN. Front Neurorobot2022;16:883816.
[41]
MontorfanoC, Giti Ruberto R. Developing cross passages and safety niches in a rationalized way using remote controlled demolition robots. In: PeilaD, ViggianiG, CelestinoT, editors. Tunnelsand underground cities. Engineering and innovation meet archaeology, architecture and art. London: CRC Press; 2020. p. 4095-103.
[42]
NguyenTH, NguyenTQ. A kinematic control algorithm for blasthole drilling robotic arm in tunneling. Sci Technol Development J2017; 20 (K5):13-22.
[43]
XieXh, ZhouL, HeQh.GRNN-based error-compensating algorithms in feeding beam of tunnel Rock-drilling robot. In:Proceedings of the 2010 IEEE International Conference on Mechatronics and Automation; 2010 Aug 4-7; Xi’an, China. Piscataway: IEEE; 2010. p. 1410-3.
[44]
LiJ, WangY, ZhangK, WangZQ, LuJX. Design and analysis of demolition robot arm based on finite element method. Adv Mech Eng2019; 11(6):1-9.
[45]
DerlukiewiczD, PtakM, WilhelmJ, JakubowskiK.The numerical- experimental studies of demolition machine operator work. In: Rusin´ ski E, Pietrusiak D, editors. Proceedings of the 13th International Scientific Conference. RESRB 2016. Lecture notes in mechanical engineering. Cham: Springer; 2017. p. 129-38.
[46]
LeeHJ, Brell-CokcanS.Reinforcement learning-based virtual fixtures for Teleoperation of Hydraulic Construction Machine. In:Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN); 2023 Aug 28-31; Busan, Republic of Korea. Piscataway: IEEE; 2023. p. 1360-7.
[47]
LeeHJ, Brell-CokcanS.Data-driven actuator model-based teleoperation assistance system. In:Proceedings of the 2023 20th International Conference on Ubiquitous Robots (UR); 2023 Jun 25-28; Honolulu, HI, USA. Piscataway: IEEE; 2023. p. 552-8.
[48]
HuangJ, CenY, ZongY, BaoJ. Random-optimal differential evolution neural network model for inverse calculation of demolition robot. In: LiuG, CenF,editors. Advances in precision instruments and optical engineering. Cham: Springer; 2022. p. 203-33.
Brokk Global. Brokk Introduces the Brokk Surface Grinder 530 [Internet]. Beijing: Brokk China; c2022 [cited 2022 Oct 19]. Available from: https://www.brokk.com/china/press-release/brokk-introduces-the-brokk-surface-grinder-530/.
[51]
HuaX, WangG, XuJ, ChenK. Reinforcement learning-based collision-free path planner for redundant robot in narrow duct. J Intell Manuf2021; 32 (2):471-82.
[52]
MnihV, KavukcuogluK, SilverD, RusuAA, VenessJ, BellemareMG, et al. Human-level control through deep reinforcement learning. Nature2015; 518 (7540):529-33.
[53]
CorucciF, RuffaldiE. Toward autonomous robots for demolitions in unstructured environments. In:MenegattiE, MichaelN, BernsK, YamaguchiH, editors. Intelligentautonomous systems. Cham: Springer; 2016. p. 1515-32.
[54]
ZongY, HuangJ, BaoJ, SunD, CenY.Target tracking method for visual autonomous localization of demolition robot. In:Proceedings of the 2022 5th World Conference on Mechanical Engineering and Intelligent Manufacturing (WCMEIM); 2022 Nov 18-20; Ma’anshan, China. Piscataway: IEEE; 2023. p. 806-10.
[55]
YeY, CenYW, XieNG.Motion control on the working device of a demolition robot based on an exponential reaching law. In:Proceedings of the IEEE 5th International Conference on Big Data and Cloud Computing; 2015 Aug 26-28; Dalian, China. Piscataway: IEEE; 2015. p. 335-7.
[56]
YoshinadaH, KurashikiK, KondoD, NagataniK, KiribayashiS, FuchidaM, et al. Dual-arm construction robot with remote-control function. In: TadokoroS, editor. Disasterrobotics. Cham: Springer; 2019. p. 195-264.
[57]
KamezakiM, IshiiH, IshidaT, SekiM, IchiryuK, KobayashiY, et al. Design of four-arm four-crawler disaster response robot. In:Proceedings of the IEEE International Conference on Robotics and Automation (ICRA); 2016 May 16-21; Stockholm, Sweden. Piscataway: IEEE; 2016. p. 2840-5.
[58]
YutaS. Development of a remotely controlled semi-underwater heavy carrier robot for unmanned construction works. J Disaster Res2017; 12 (3):432-45.
[59]
OotsuboK, KatoD, KawamuraT, YamadaH.Support system for teleoperation of slope shaping by a construction robot. In:Proceedings of the 8th IEEE/SICE International Symposium on System Integration (SII); 2015 Dec 11-13; Nagoya, Japan. Piscataway: IEEE; 2016. p. 918-23.
[60]
OotsuboK, KawamuraT, YamadaH. Construction tele-robotics system with AR presentation. J Phys Conf Ser2013; 433(1):012029.
[61]
NubertJ, KhattakS, HutterM.Graph-based multi-sensor fusion for consistent localization of autonomous construction robots. In:Proceedings of the 2022 International Conference on Robotics and Automation (ICRA); 2022 May 23-27; Philadelphia, PA, USA. Piscataway: IEEE; 2022. p. 10048-54.
[62]
FukuiR, KatoY, KanayamaG, TakahashiR, NakaoM. Construction planning for a modularized rail structure:type selection of rail structure modules and dispatch planning of constructor robots. In: GroßR, KollingA, BermanS, FrazzoliE, MartinoliA, MatsunoF, et al. editors. Distributed autonomous robotic systems. Cham: Springer; 2018. p. 605-17.
[63]
FukuiR, KatoY, TakahashiR, WanWW, NakaoM. Automated construction system of robot locomotion and operation platform for hazardous environments—basic system design and feasibility study of module transferring and connecting motions. J Field Robot2016; 33(6):751-64.
[64]
YasudaM, FukuiR, WarisawaS. Stable, sensor-less and compliance-less module connection for automated construction system of a modularized rail structure. In: Proceedings of the 2012 IEEE International Conference on Robotics and Automation (ICRA); 2021 May 30-Jun 5; Xi’an, China. Piscataway: IEEE; 2021. p. 6948-54.
[65]
HeDC, LiY, MengXP, SiQR. Anti-slip control for unmanned underwater tracked bulldozer based on active disturbance rejection control. Mechatronics2022;84:102803.
[66]
MinamotoM, KawashimaK, KannoT.Effect of force feedback on a bulldozer- type robot. In:Proceedings of the 2016 IEEE International Conference on Mechatronics and Automation; 2016 Aug 7-10; Harbin, China. Piscataway: IEEE; 2016. p. 2203-8.
[67]
WestC, ArvinF, CheahW, WestA, WatsonS, GiulianiM, et al. A debris clearance robot for extreme environments. In: AlthoeferK, KonstantinovaJ, ZhangK, editors. Towardsautonomous robotic systems. Cham: Springer; 2019. p. 148-59.
[68]
KamataM, YamazakiS, TaniseY, YamadaY, NakamuraT. Morphological change in peristaltic crawling motion of a narrow pipe inspection robot inspired by earthworm’s locomotion. Adv Robot2018; 32(7):386-97.
[69]
GaoH, DuJ, TangM, ShiW.Research on a new type peristaltic micro in-pipe robot. In:Proceedings of the 2011 IEEE/ICME International Conference on Complex Medical Engineering; 2011 May 22-25; Harbin, China. Piscataway: IEEE; 2011. p. 26-30.
[70]
RazakAAA, AbdullahAH, KamarudinK, SaadFSA, ShukorSA, MustafaH, et al. Mobile robot structure design, modeling and simulation for confined space application. In:Proceedings of the 2nd IEEE International Symposium on Robotics and Manufacturing Automation (ROMA); 2016 Sep 25-27; Ipoh, Malaysia. Piscataway: IEEE; 2017. p. 1-5.
[71]
MateosLA, ZhouK, VinczeM. Towards efficient pipe maintenance:DeWaLoP in-pipe robot stability controller. In:Proceedings of the 2012 IEEE International Conference on Mechatronics and Automation; 2012 Aug 5-8; Chengdu, China. Piscataway: IEEE; 2012. p. 1-6.
[72]
LiuF, LiuW, LuoH. Operational stability control of a buried pipeline maintenance robot using an improved PSO-PID controller. Tunn Undergr Space Technol2023;138:105178.
[73]
JangH, KimHM, LeeMS, SongYH, LeeY, RyewWR, et al. Development of modularized in-pipe inspection robotic system: MRINSPECT VII+. Robotica2022; 40(5):1361-84.
[74]
HuaX, WangG, ZhangS, LiuX, ChenK. Trajectory planning of redundant robot for painting inner surface of complex duct considering obstacle avoidance. Robot2019; 41(5):690-6.
[75]
ZhengT, WangX, LiH, ZhaoCA, JiangZH, HuangQ, et al. Design of a robot for inspecting the multishape pipeline systems. IEEE/ASME Trans Mechatron2022; 27(6):4608-18.
[76]
HeKM, SunJ, TangXO.Single image haze removal using dark channel prior. In:Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition; 2009 Jun 20-25; Miami, FL, USA. Piscataway: IEEE; 2009. p. 1956-63.
LiY, DuthonP, ColombM, Ibanez-GuzmanJ. What happens for a ToF LiDAR in fog? IEEE Trans Intell Transp Syst 2021; 22(11):6670-81.
[79]
EmterT, FreseC, ZubeA, PetereitJ. Algorithm toolbox for autonomous mobile robotic systems. ATZoffhighway worldw2017; 10(3):48-53.
[80]
LinXB, SongD, QinM, ZhangWT, HeXY, XieB.An automatic tunnel shotcrete robot. In:Proceedings of the 2019 Chinese Automation Congress (CAC); 2019 Nov 22-24; Hangzhou, China. Piscataway: IEEE; 2020. p. 3858-63.
[81]
OuyangQ, LinYH, ZhangXL, FanYX, YangWJ, HuangT.Application of 3D reconstruction technology based on an improved MC algorithm in a shotcreting robot. Appl Opt 2022; 61(29):8649-56.
[82]
ZhouY, TuzelO.VoxelNet:end-to-end learning for point cloud based 3D object detection. In:Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition; 2018 Jun 18-23; Salt Lake City, UT, USA. Piscataway: IEEE; 2018. p. 4490-9.
[83]
WangYY, LiXC, ZhaoXL, HuY, TangGW, BuBL. Performance of the micro- clean space system as a personal protection method at coal mine excavation face. Powder Technol2024;437:119517.
[84]
KrusePW, CunliffeG, WilsonJ. Use of remotely operated excavators for environmental restoration projects. In:Proceedings of the Canadian Nuclear Society 15. Annual Conference; 1994 Jun 5-8; Montreal, PQ, Canada.1994. p. 9.
[85]
WoockP, PetereitJ, FreyC, BeyererJ. ROBDEKON—competence center for decontamination robotics. Automatisierungstechnik2022; 70(10):827-37.
[86]
HeideNF, PetereitJ. Machine learning for the perception of autonomous construction machinery. Automatisierungstechnik2023; 71(3):219-32.
[87]
BirdB, NancekievillM, WestA, HaymanJ, BallardC, JonesW, et al. Vega—a small, low cost, ground robot for nuclear decommissioning. J Field Robot2022; 39(3):232-45.
[88]
BurrellT, MontazeriA, MonkS, TaylorCJ. Feedback control-based inverse kinematics solvers for a nuclear decommissioning robot. IFAC-PapersOnLine2016; 49(21):177-84.
[89]
LaiYC, SmithS. Metaheuristic minimum dose path planning for nuclear power plant decommissioning. Ann Nucl Energy2022;166:108800.
[90]
ChaoN, LiuYK, XiaH, PengMJ, AyodejiA. DL-RRT* algorithm for least dose path re-planning in dynamic radioactive environments. Nucl Eng Technol2019; 51(3):825-36.
[91]
TsitsimpelisI, WestA, LicataM, AspinallMD, JazbecA, SnojL, et al. Simultaneous, robot-compatible c-Ray spectroscopy and imaging of an operating nuclear reactor. IEEE Sens J2021; 21(4):5434-43.
[92]
BakariMJ, ZiedKM, SewardDW. Development of a multi-arm mobile robot for nuclear decommissioning tasks. Int J Adv Robot Syst2007; 4 (4):387-406.
[93]
LiCQ, KhanH, LeeJ, KimJ, LeeMC. Fuzzy TSMCSPO for trajectory tracking of nuclear reactor dismantlement robot manipulator. IEEE Access2023;11:38696-707.
[94]
LuoY, TaoJG, SunQJ, DengLP, DengZQ.A new underwater robot for crack welding in nuclear power plants. In:Proceedings of the 2018 IEEE International Conference on Robotics and Biomimetics (ROBIO); 2018 Dec 12-15; Kuala Lumpur, Malaysia. Piscataway: IEEE; 2019. p. 77-82.
[95]
QinWB, ChenEJ, WangF, LiuWL, ZhouC. Data-driven models in reliability analysis for tunnel structure: a systematic review. Tunn Undergr Space Technol2024;152:105928.
[96]
ArmaghaniDJ, MohamadET, NarayanasamyMS, NaritaN, YagizS. Development of hybrid intelligent models for predicting TBM penetration rate in hard rock condition. Tunn Undergr Space Technol2017;63:29-43.
[97]
GongMD, ZhaoDX, FengSZ, WeiHL, YamadaH.Force feedback model of electro-hydraulic servo tele-operation robot based on velocity control. In:Proceedings of the 2008 IEEE Conference on Robotics, Automation and Mechatronics; 2008 Sep 21-24; Chengdu, China. Piscataway: IEEE; 2008. p. 912-5.
[98]
HommaK, AraiT, AdachiH, NakamuraT. Research on components of underground excavation robot. In:WatsonGH, TuckerRL, WaltersJK, editors. Automationand robotics in construction. Amsterdam: Elsevier; 1993. p. 245-52.
[99]
ZhangQL, ZhuYW, MaR, DuSL, ShaoK, JingLJ, et al. Intelligent tunnelling robot system for deep-buried long tunnels. Front Earth Sci2023;11:1135948.
[100]
ZhouC, WangY, LiR, GuanT, LiuZ, PengG, et al. Artificial intelligence technology for path planning of automated earthwork machinery. J Field Robot2024;42:1887-913.
[101]
ZhangDP, ZhaoBW, ZhuKQ, JiangHY. Dynamic response of deep-sea trawl system during towing process. J Mar Sci Eng2023; 11(1):145.
[102]
FreudenthalT, WeferG. Shallow drilling in the deep sea:the sea floor drill rig MeBo. In:Proceedings of the OCEANS 2009-EUROPE; 2009 May 11-14; Bremen, Germany. Piscataway: IEEE; 2009. p. 1-4.
[103]
Sverdrup-ThygesonJ, KelasidiE, PettersenKY, GravdahlJT. The underwater swimming manipulator—a bio-inspired AUV. In:Proceedings of the 2016IEEE/OES Autonomous Underwater Vehicles (AUV); 2016 Nov 6-9; Tokyo, Japan. Piscataway: IEEE; 2016. p. 387-95.
[104]
TadamiN, NagaiM, NakatakeT, FujiwaraA, YamadaY, NakamuraT, et al. Curved excavation by a sub-seafloor excavation robot. In:Proceedings of the 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS); 2017 Sep 24-28; Vancouver, BC, Canada. Piscataway: IEEE; 2017. p. 4950-6.
[105]
NagaiM, MizushinaA, NakamuraT, SugimotoF, WatariK, NakajoH, et al. Development of a hydraulic artificial muscle for a deep-seafloor excavation robot with a peristaltic crawling mechanism. In: LiuHH, KubotaN, ZhuXY, DillmannR, ZhouDL, editors. Intelligentrobotics and applications. Cham: Springer; 2015. p. 379-89.
[106]
MurphyRR. Designing a robot to recover a sunken submarine is hard. Sci Robot2023; 8(81):eadj8287.
[107]
KimTS, JangIS, ShinCJ, LeeMK.Underwater construction robot for rubble leveling on the seabed for port construction. In:Proceedings of the 2014 14th International Conference on Control, Automation and Systems (ICCAS2014); 2014 Oct 22-25; Gyeonggi-do, Republic of korea. Piscataway: IEEE; 2014. p. 1657-61.
[108]
KimTS, KimCH, LeeMK. Study on the design and the control of an underwater construction robot for port construction. J Korean Navig Port Reserch2015; 39(3):253-60.
[109]
SongY, ChoiSJ.Underwater 3D reconstruction for underwater construction robot based on 2D multibeam imaging sonar. J Ocean Eng Technol 2016; 30 (3):227-33.
[110]
AncutiTCO, AncutiC, BaltaTH.Effective local backscattering estimation for underwater image enhancement. In: Proceedings of the OCEANS 2018 MTS/ IEEE Charleston; 2018 Oct 22-25; Charleston, SC, USA. Piscataway: IEEE; 2019. p. 1-4.
[111]
SitorusPE, KoJH, KwonOS. Parameter study of chain trenching machines of underwater construction robots via analytical model. In: Proceedings of the OCEANS 2016 MTS/IEEE Monterey; 2016 Sep 19-23; Monterey, CA, USA. Piscataway: IEEE; 2016. p. 1-6.
[112]
HongSM, KangHS, ChoiHS, KimJY.Development of the control algorithm for longitudinal motion of underwater construction robot with trenching. In: Proceedings of the 2017 IEEE Underwater Technology (UT); 2017 Feb 21-24; Busan, Republic of korea. Piscataway: IEEE; 2017. p. 1-5.
[113]
McIsaacKA, OstrowskiJP. A geometric approach to anguilliform locomotion:modelling of an underwater eel robot. In:Proceedings of the Proceedings 1999 IEEE International Conference on Robotics and Automation (Cat. No. 99CH36288C); 1999 May 10-15; Detroit, MI, USA. Piscataway: IEEE; 2002 p. 2843-8.
[114]
GormaW, PostMA, WhiteJ, GardnerJ, LuoY, KimJ, et al. Development of modular bio-inspired autonomous underwater vehicle for close subsea asset inspection. Appl Sci2021; 11(12):5401.
KimH, ChoiY, ParkJ, LeeJ, LeeJ, LeeH, et al. Active control strategy for trenching work of track-based underwater construction robot. In:Proceedings of the Thirteenth ISOPE Pacific/Asia Offshore Mechanics Symposium; 2018 Oct 14-17; Jeju, Republic of Korea; Richardson, TX, OnePetro; 2018. p. ISOPE-P-18-118.
[117]
FinnC, AbbeelP, LevineS. Model-agnostic meta-learning for fast adaptation of deep networks. 2017. arXiv:1703.03400.
[118]
MoonJ, BaeSH, CashmoreM.Meta reinforcement learning based underwater manipulator control. In:Proceedings of the 2021 21st International Conference on Control, Automation and Systems (ICCAS); 2021 Oct 12-15; Jeju, Republic of korea. Piscataway: IEEE; 2021. p. 1473-6.
[119]
WeiH, ZhangY, ZhangT, GuanY, XuK, DingX, et al. Review on bioinspired planetary regolith-burrowing robots. Space Sci Rev2021;217:87.
[120]
LordosG, BrownMJ, LatyshevK, LiaoA, ShahS, MezaC, et al. WORMS: field- reconfigurable robots for extreme lunar terrain. In: Proceedings of the 2023 IEEE Aerospace Conference; 2023 Mar 4-11; Big Sky, MT, USA. Piscataway: IEEE; 2023. p. 1-21.
[121]
YaziciS. Building in extraterrestrial environments: T-brick shell. J Archit Eng2018; 24(1):04017037.
[122]
ChengS, GaoY, ShiY, XiaoL, DingL, ZhouC, et al. Advances in in-situ resources utilization for extraterrestrial construction. Adv Space Res2024; 74 (7):3297-325.
[123]
ChenQY, GaoYY, DingLY, ZhouC, HanWB, ZhouY, et al. Genetic algorithm- based multiobjective optimization for 3D printable design of a double-shell lunar habitat structure. J Aerosp Eng2023; 36(6):04023069.
[124]
ZhouC, WangYX, LuYA, YouK, JiangYT, WuZA. Transformer-based berm detection for automated bulldozer safety in edge dumping. Autom Constr2024;166:105642.
[125]
SunZY, YangH, DongQ, MoY, LiH, JiangZH.Autonomous assembly method of 3-arm robot to fix the multipin and hole load plate on a space station. Space 2021;2021:9815389.
[126]
ZhouC, ChenR, XuJ, DingLY, LuoHB, FanJ, et al. In-situ construction method for lunar habitation: Chinese super mason. Autom Constr2019;104:66-79.
[127]
TakahashiO, SchillingRJ. Motion planning in a plane using generalized Voronoi diagrams. IEEE Trans Robot Autom1989; 5(2):143-50.
[128]
LaValleS.Rapidly-exploring random trees: a new tool for path planning. Annu Res Rep 1989.
[129]
YueC, LinT, ZhangX, ChenX, CaoX. Hierarchical path planning for multi-arm spacecraft with general translational and rotational locomotion mode. Sci China Technol Sci2023; 66(4):1180-91.
[130]
LiDL, ZhongL, ZhuW, XuZP, TangQR, ZhanWH. A survey of space robotic technologies for on-orbit assembly. Space Sci Technol2022;2022:9849170.
[131]
BoningP, DubowskyS. Coordinated control of space robot teams for the on- orbit construction of large flexible space structures. Adv Robot2010; 24 (3):303-23.
[132]
YangXX, GeSS, HeW. Dynamic modelling and adaptive robust tracking control of a space robot with two-link flexible manipulators under unknown disturbances. Int J Control2018; 91(4):969-88.
[133]
GalardiniDG, KapellosK, MaesenE.Vision & interactive autonomy bi-lateral experiments on the Japanese satellite ETS-VII. In:Proceedings of the fifth International Symposium on Artificial intelligence, Robotics and Automation in Space; 1999 Jun 1-3, Noordwijk:Netherlands. Paris: European Space Agency, 1999. p. 217-24.
[134]
MuellerRP, KingRH. Trade study of excavation tools and equipment for lunar outpost development and ISRU. AIP Conf Proc2008; 969(1):237-44.
[135]
MelenbrinkN, WerfelJ, MengesA. On-site autonomous construction robots: towards unsupervised building. Autom Constr2020;119:130012.
[136]
SchusterMJ, BrunnerSG, BussmannK, BüttnerS, DömelA, HellererM, et al. Towards autonomous planetary exploration. J Intell Robot Syst2019; 93(3- 4):461-94.
[137]
CloudJM, TramMQ, BeksiWJ, DuPuisMA.Lunar excavator mission operations using dynamic movement primitives. In:Proceedings of the 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS); 2023 Oct 1-5; Detroit, MI, USA. Piscataway: IEEE; 2023. p. 10708-15.
[138]
ZhouC, GaoYY, ZhouY, SheW, ShiYS, DingLY, et al. Properties and characteristics of regolith-based materials for extraterrestrial construction. Engineering2024;37:159-81.
[139]
ZacharakiA, KostavelisI, GasteratosA, DokasI. Safety bounds in human robot interaction: a survey. Saf Sci2020;127:104667.
[140]
KimD, LiuMY, LeeS, KamatVR. Remote proximity monitoring between mobile construction resources using camera-mounted UAVs. Autom Constr2019;99:168-82.
[141]
YouK, DingLY, ZhouC, DouQL, WangXP, HuB.5G-based earthwork monitoring system for an unmanned bulldozer. Autom Constr 2021;131:103891.
[142]
FazeliN, OllerM, WuJ, WuZ, TenenbaumJB, RodriguezA. See, feel, act: hierarchical learning for complex manipulation skills with multisensory fusion. Sci Robot2019; 4(26):eaav3123.
[143]
YuFW, WuYJ, MaSC, XuMK, LiHY, QuHY, et al. Brain-inspired multimodal hybrid neural network for robot place recognition. Sci Robot2023; 8(78): eabm6996.
[144]
BruckJN, WalmsleySF, JanikVM. Cross-modal perception of identity by sound and taste in bottlenose dolphins. Sci Adv2022; 8(20):eabm7684.
[145]
MagañaOAV, BarasuolV, CamurriM, FranceschiL, FocchiM, PontilM, et al. Fast and continuous foothold adaptation for dynamic locomotion through CNNs. IEEE Robot Autom Lett2019; 4(2):2140-7.
[146]
YinH, VaravaA, KragicD. Modelinglearning, perception, and control methods for deformable object manipulation. Sci Robot2021; 6(54): eabd8803.
[147]
HwangboJ, LeeJ, DosovitskiyA, BellicosoD, TsounisV, KoltunV, et al. Learning agile and dynamic motor skills for legged robots. Sci Robot2019; 4 (26):eaau5872.
LuYA, YouK, ZhouC, ChenJX, WuZA, JiangYT, et al. Video surveillance-based multi-task learning with swin transformer for earthwork activity classification. Eng Appl Artif Intell2024;131:107814.
PDF
(5645KB)
