Snake robots can move flexibly due to their special bodies and gaits. However, it is difficult to plan their motion in multi-obstacle environments due to their complex models. To solve this problem, this work investigates a reinforcement learning-based motion planning method. To plan feasible paths, together with a modified deep Q-learning algorithm, a Floyd-moving average algorithm is proposed to ensure smoothness and adaptability of paths for snake robots’ passing. An improved path integral algorithm is used to work out gait parameters to control snake robots to move along the planned paths. To speed up the training of parameters, a strategy combining serial training, parallel training, and experience replaying modules is designed. Moreover, we have designed a motion planning framework consists of path planning, path smoothing, and motion planning. Various simulations are conducted to validate the effectiveness of the proposed algorithms.

Motivated by practical applications of inspection and maintenance, we have developed a wall-climbing robot with passive compliant mechanisms that can autonomously adapt to curved surfaces. At first, this paper presents two failure modes of the traditional wall-climbing robot on the variable curvature wall surface and further introduces the designed passive compliant wall-climbing robot in detail. Then, the motion mechanism of the passive compliant wall-climbing robot on the curved surface is analyzed from stable adsorption conditions, parameter design process, and force analysis. At last, a series of experiments have been carried out on load capability and curved surface adaptability based on a developed principle prototype. The experimental results indicated that the wall-climbing robot with passive compliant mechanisms can effectively promote both adsorption stability and adaptability to variable curvatures.

The loading and unloading operations of smart logistic application robots depend largely on their perception system. However, there is a paucity of study on the evaluation of Lidar maps and their SLAM algorithms in complex environment navigation system. In the proposed work, the Lidar information is finetuned using binary occupancy grid approach and implemented Improved Self-Adaptive Learning Particle Swarm Optimization (ISALPSO) algorithm for path prediction. The approach makes use of 2D Lidar mapping to determine the most efficient route for a mobile robot in logistical applications. The Hector SLAM method is used in the Robot Operating System (ROS) platform to implement mobile robot real-time location and map building, which is subsequently transformed into a binary occupancy grid. To show the path navigation findings of the proposed methodologies, a navigational model has been created in the MATLAB 2D virtual environment using 2D Lidar mapping point data. The ISALPSO algorithm adapts its parameters inertia weight, acceleration coefficients, learning coefficients, mutation factor, and swarm size, based on the performance of the generated path. In comparison to the other five PSO variants, the ISALPSO algorithm has a considerably shorter path, a quick convergence rate, and requires less time to compute the distance between the locations of transporting and unloading environments, based on the simulation results that was generated and its validation using a 2D Lidar environment. The efficiency and effectiveness of path planning for mobile robots in logistic applications are validated using Quanser hardware interfaced with 2D Lidar and operated in environment 3 using proposed algorithm for production of optimal path.

In this paper, we consider the problem of contact parameters (slippage and sinkage) estimation for multi-modal robot locomotion on granular terrains. To describe the contact events in the same framework for robots operated at different modes (e.g., wheel, leg), we propose a unified description of contact parameters for multi-modal robots. We also provide a parameter estimation method for multi-modal robots based on CNN and DWT (discrete wavelet transformation) techniques and verify its effectiveness over different types of granular terrains. Besides motion modes, this paper also considers the influence of slope angles and the robot’s handing angles over contact parameters. Through comparison and analysis of the prediction results, our method can not only effectively predict the contact parameters of multi-modal robot locomotion on a granular medium (better than $96\%$ accuracy) but also achieves the same or better performance when compared to other (direct) contact measurement methods designed for individual motion modes, that is, single-modal robots such as quadruped robots and mars rovers. Our proposed unified contact parameter estimation method can be useful for studying the interaction mechanics between multi-modal robots and granular terrains as well as terrain classification tasks due to its superior sensitivity which is analyzed in the experiments.

Currently, workers in sand casting face harsh environments and the operation safety is poor. Existing pouring robots have insufficient stability and load-bearing capacity and cannot perform intelligent pouring according to the demand of pouring process. In this paper, a hybrid pouring robot is proposed to solve these limitations, and a vision-based hardware-in-the-loop (HIL) control technology is designed to achieve the real-time control problems of simulated pouring and pouring process. Firstly, based on the pouring mechanism and the motion demand of ladle, a hybrid pouring robot with a 2UPR-2RPU parallel mechanism as the main body is designed. And the equivalent hybrid kinematic model was established by using Eulerian method and differential motion. Subsequently, a motion control strategy based on HIL simulation technique was designed and presented. The working space of the robot was obtained through simulation experiments to meet the usage requirements. And the stability of the robot was tested through the key motion parameters of the robot joints. Based on the analysis of pouring quality and trajectory, optimal dynamic parameters for the experimental prototype are obtained through water simulation experiments, the pouring liquid height area is 35–40 cm, the average flow rate of pouring liquid is 112 cm3/s, and the ladle tilting speed is 0.0182 rad/s. Experimental results validate the reasonableness of the designed pouring robot structure. Its control system realizes the coordinated movement of each branch chain to complete the pouring tasks with different variable parameters. Consequently, the designed pouring robot will significantly enhance the automation level of the casting industry.

SLAM Benchmark plays a pivotal role in the field by providing a common ground for performance evaluation. In this paper, a novel methodology of simultaneous localization and mapping benchmark and map accuracy improvement (SLAMB&MAI) is introduced. It can objectively evaluate errors of localization and mapping, and further improve map accuracy by utilizing evaluation results as feedback. The proposed benchmark transforms all elements into a global frame and measures the errors between them. The comprehensiveness consists in the benchmark of both localization and mapping, and the objectivity consists in the consideration of the correlation between localization and mapping by the preservation of the original pose relations between all reference frames. The map accuracy improvement is realized by first obtaining the optimization that minimizes the errors between the estimated trajectory and ground truth trajectory and then applying it to the estimated map. The experimental results showed that the map accuracy can be improved by an average of 15%. The optimization that yields minimal localization errors is obtained by the proposed Centre Point Registration-Iterative Closest Point (CPR-ICP). This proposed Iterative Closest Point (ICP) variant pre-aligns two point clouds by their centroids and least square planes and then uses traditional ICP to minimize the error between them. The experimental results showed that CPR-ICP outperformed traditional ICP, especially in cases involving large-scale environments. To the extent of our knowledge, this is the first work that can not only objectively benchmark both localization and mapping but also revise the estimated map and increase its accuracy, which provides insights into the acquisition of ground truth map and robot navigation.

This paper presents a comprehensive strategy to improve the locomotion performance of humanoid robots on various slippery floors. The strategy involves the implementation and adaptation of a divergent component of motion (DCM) based control architecture for the humanoid NAO, and the introduction of an embedded yaw controller (EYC), which is based on a proportional-integral-derivative (PID) control algorithm. The EYC is designed not only to address the slip behavior of the robot on low-friction floors but also to tackle the issue of non-straight walking patterns that we observed in this humanoid, even on non-slippery floors. To fine-tune the PID gains for the EYC, a systematic trial-and-error approach is employed. We iteratively adjusted the P (Proportional), I (Integral), and D (Derivative) parameters while keeping the others fixed. This process allowed us to optimize the PID controller’s response to different walking conditions and floor types. A series of locomotion experiments are conducted in a simulated environment, where the humanoid step frequency and PID gains are varied for each type of floor. The effectiveness of the strategy is evaluated using metrics such as robot stability, energy consumption, and task duration. The results of the study demonstrate that the proposed approach significantly improves humanoid locomotion on different slippery floors, by enhancing stability and reducing energy consumption. The study has practical implications for designing more versatile and effective solutions for humanoid locomotion on challenging surfaces and highlights the adaptability of the existing controller for different humanoid robots.

Reinforcement learning (RL) has been successfully applied to a wealth of robot manipulation tasks and continuous control problems. However, it is still limited to industrial applications and suffers from three major challenges: sample inefficiency, real data collection, and the gap between simulator and reality. In this paper, we focus on the practical application of RL for robot assembly in the real world. We apply enlightenment learning to improve the proximal policy optimization, an on-policy model-free actor-critic reinforcement learning algorithm, to train an agent in Cartesian space using the proprioceptive information. We introduce enlightenment learning incorporated via pretraining, which is beneficial to reduce the cost of policy training and improve the effectiveness of the policy. A human-like assembly trajectory is generated through a two-step method with segmenting objects by locations and iterative closest point for pretraining. We also design a sim-to-real controller to correct the error while transferring to reality. We set up the environment in the MuJoCo simulator and demonstrated the proposed method on the recently established The National Institute of Standards and Technology (NIST) gear assembly benchmark. The paper introduces a unique framework that enables a robot to learn assembly tasks efficiently using limited real-world samples by leveraging simulations and visual demonstrations. The comparative experiment results indicate that our approach surpasses other baseline methods in terms of training speed, success rate, and efficiency.

Real-time and efficient path planning is critical for all robotic systems. In particular, it is of greater importance for industrial robots since the overall planning and execution time directly impact the cycle time and automation economics in production lines. While the problem may not be complex in static environments, classical approaches are inefficient in high-dimensional environments in terms of planning time and optimality. Collision checking poses another challenge in obtaining a real-time solution for path planning in complex environments. To address these issues, we propose an end-to-end learning-based framework viz., Path Planning and Collision checking Network (PPCNet). The PPCNet generates the path by computing waypoints sequentially using two networks: the first network generates a waypoint, and the second one determines whether the waypoint is on a collision-free segment of the path. The end-to-end training process is based on imitation learning that uses data aggregation from the experience of an expert planner to train the two networks, simultaneously. We utilize two approaches for training a network that efficiently approximates the exact geometrical collision checking function. Finally, the PPCNet is evaluated in two different simulation environments and a practical implementation on a robotic arm for a bin-picking application. Compared to the state-of-the-art path-planning methods, our results show significant improvement in performance by greatly reducing the planning time with comparable success rates and path lengths.

Physically compliant actuator brings significant benefits to robots in terms of environmental adaptability, human–robot interaction, and energy efficiency as the introduction of the inherent compliance. However, this inherent compliance also limits the force and position control performance of the actuator system due to the induced oscillations and decreased mechanical bandwidth. To solve this problem, we first investigate the dynamic effects of implementing variable physical damping into a compliant actuator. Following this, we propose a structural scheme that integrates a variable damping element in parallel to a conventional series elastic actuator. A damping regulation algorithm is then developed for the parallel spring-damping actuator (PSDA) to tune the dynamic performance of the system while remaining sufficient compliance. Experimental results show that the PSDA offers better stability and dynamic capability in the force and position control by generating appropriate damping levels.

In this study, a fuzzy reinforcement learning control (FRLC) is proposed to achieve trajectory tracking of a differential drive mobile robot (DDMR). The proposed FRLC approach designs fuzzy membership functions to fuzzify the relative position and heading between the current position and a prescribed trajectory. Instead of fuzzy inference rules, the relationship between the fuzzy inputs and actuator voltage outputs is built using a reinforcement learning (RL) agent. Herein, the deep deterministic policy gradient (DDPG) methodology consisted of actor and critic neural networks is employed in the RL agent. Simulations are conducted with considering varying slip ratio disturbances, different initial positions, and two different trajectories in the testing environment. In the meantime, a comparison with the classical DDPG model is presented. The results show that the proposed FRLC is capable of successfully tracking different trajectories under varying slip ratio disturbances as well as having performance superiority to the classical DDPG model. Moreover, experimental results validate that the proposed FRLC is also applicable to real mobile robots.

A collision-free path planning method is proposed based on learning from demonstration (LfD) to address the challenges of cumbersome manual teaching operations caused by complex action of yarn storage, variable mechanism positions, and limited workspace in preform weaving. First, by utilizing extreme learning machines (ELM) to autonomously learn the teaching data of yarn storage, the mapping relationship between the starting and ending points and the teaching path points is constructed to obtain the imitation path with similar storage actions under the starting and ending points of the new task. Second, an improved rapidly expanding random trees (IRRT) method with adaptive direction and step size is proposed to expand path points with high quality. Finally, taking the spatical guidance point of imitation path as the target direction of IRRT, the expansion direction is biased toward the imitation path to obtain a collision-free path that meets the action yarn storage. The results of different yarn storage examples show that the ELM-IRRT method can plan the yarn storage path within 2s–5s when the position of the mechanism changes in narrow spaces, avoiding tedious manual operations that program the robot movements, which is feasible and effective.

Accurate prediction for mechanisms’ dynamic responses has always been a challenging task for designers. For modeling easiness purposes, mechanisms’ synthesis and optimization have been mostly limited to rigid systems, making consequently the designer unable to vow that the manufactured mechanism satisfies the target responses. To address this limitation, flexible mechanism synthesis is aimed in this work. Two benchmark mechanisms being the core of myriad mechanical devices are of scope, mainly, the flexible slider-crank and the four-bar. In addition to the mechanism dimensions, materials properties have been embedded in the synthesis problem. Two responses are of interest for the slider-crank mechanism, the slider velocity, and the midpoint axial displacement for the flexible connecting rod. Whereas five responses have been compiled for the four-bar mechanism synthesis. A comparative analysis of seven optimization techniques to solve the synthesis problem for both mechanisms has been performed. Subsequently, an executable computer-aided design tool for mechanisms synthesis has been developed under MATLAB®. Numerical outcomes emphasize the limits of a single-response-based synthesis for a flexible mechanism. It has been proven that combining different responses alleviates possible error and fulfill high-accuracy requirement.

Safe and socially compliant navigation in a crowded environment is essential for social robots. Numerous research efforts have shown the advantages of deep reinforcement learning techniques in training efficient policies, while most of them ignore fast-moving pedestrians in the crowd. In this paper, we present a novel design of safety measure, named Risk-Area, considering collision theory and motion characteristics of different robots and humans. The geometry of Risk-Area is formed based on the real-time relative positions and velocities of the agents in the environment. Our approach perceives risk in the environment and encourages the robot to take safe and socially compliant navigation behaviors. The proposed method is verified with three existing well-known deep reinforcement learning models in densely populated environments. Experiment results demonstrate that our approach combined with the reinforcement learning techniques can efficiently perceive risk in the environment and navigate the robot with high safety in the crowds with fast-moving pedestrians.

The recognizing underwater targets is a crucial component of autonomous underwater vehicle patrols and detection efforts. In the process of visual image recognition in real underwater environment, the spatial and semantic features of the target often appear to different degrees of loss, and the scarcity of specific types of underwater samples leads to unbalanced data on categories. This kind of problem makes the target features appear weak and seriously affects the accuracy of underwater target recognition. Traditional deep learning methods based on data and feature enhancement cannot achieve ideal recognition effect. Based on the above difficulties, this paper proposes an improved feature enhancement network for weak feature target recognition. Firstly, a multi-scale spatial and semantic feature enhancement module is constructed to extract the feature information of the extraction target accurately. Secondly, this paper solves the influence of target feature distortion on classification through multi-scale feature comparison of positive and negative samples. Finally, the Rank & Sort Loss function was used to train the depth target detection to solve the problem of recognition accuracy under highly unbalanced sample data. Experimental results show that the recognition accuracy of the proposed method is 2.28% and 3.84% higher than that of the existing algorithms in the recognition of underwater fuzzy and distorted target images, which demonstrates the effectiveness and superiority of the proposed method.

The current LiDAR-inertial odometry is prone to cumulative Z-axis error when it runs for a long time. This error can easily lead to the failure to detect the loop-closing in the correct scenario. In this paper, a ground-constrained LiDAR-inertial SLAM is proposed to solve this problem. Reasonable constraints on the ground motion of the mobile robot are incorporated to limit the Z-axis drift error. At the same time, considering the influence of initial positioning error on navigation, a keyframe selection strategy is designed to effectively improve the flatness and accuracy of positioning and the efficiency of loop detection. If GNSS is available, the GNSS factor is added to eliminate the cumulative error of the trajectory. Finally, a large number of experiments are carried out on the self-developed robot platform to verify the effectiveness of the algorithm. The results show that this method can effectively improve location accuracy in outdoor environments, especially in environments of feature degradation and large scale.

Differentially flat under-actuated robots are characterized by more degrees of freedom (DOF) than actuators: this makes possible the design of lightweight cheap robots with high dexterity. The main issue of such robots is the control of the passive joint, which requires accurate dynamic modeling of the robot.

Friction is usually discarded to simplify the models, especially in the case of low-speed trajectories. However, this simplification leads to oscillations of the end-effector about the final position, which are incompatible with fast and accurate motions.

This paper focuses on planar $n$-DOF serial robotic arms with $n-1$ actuated rotational joints plus one final passive rotational joint with stiffness and friction properties. These robots, if properly balanced, are differentially flat. When the non-actuated joint can be considered frictionless, differentially flat robots can be controlled in open loop, calculating the motor torques demanded by point-to-point motions. This paper extends the open-loop control to robots with a passive joint with viscous friction adopting a Laplace transform method. This method can be adopted by exploiting the particular structure of the equations of motion of differentially flat under-actuated robots in which the last equations are linear. Analytical expressions of the motor torques are obtained. The work is enriched by an experimental validation of a $2$-DOF under-actuated robot and by numerical simulations of the $2$- and $4$-DOF robots showing the suppression of unwanted oscillations.

Autonomous fabric manipulation is a challenging task due to complex dynamics and potential self-occlusion during fabric handling. An intuitive method of fabric-folding manipulation first involves obtaining a smooth and unfolded fabric configuration before the folding process begins. However, the combination of quasi-static actions like pick & place and dynamic action like fling proves inadequate in effectively unfolding long-sleeved T-shirts with sleeves mostly tucked inside the garment. To address this limitation, this paper introduces an enhanced quasi-static action called pick & drag, specifically designed to handle this type of fabric configuration. Additionally, an efficient dual-arm manipulation system is designed in this paper, which combines quasi-static (including pick & place and pick & drag) and dynamic fling actions to flexibly manipulate fabrics into unfolded and smooth configurations. Subsequently, once it is confirmed that the fabric is sufficiently unfolded and all fabric keypoints are detected, the keypoint-based heuristic folding algorithm is employed for the fabric-folding process. To address the scarcity of publicly available keypoint detection datasets for real fabric, we gathered images of various fabric configurations and types in real scenes to create a comprehensive keypoint dataset for fabric folding. This dataset aims to enhance the success rate of keypoint detection. Moreover, we evaluate the effectiveness of our proposed system in real-world settings, where it consistently and reliably unfolds and folds various types of fabrics, including challenging situations such as long-sleeved T-shirts with most parts of sleeves tucked inside the garment. Specifically, our method achieves a coverage rate of 0.822 and a success rate of 0.88 for long-sleeved T-shirts folding. Supplemental materials and dataset are available on our project webpage at https://sites.google.com/view/fabricfolding.

In the pipeline industry, it is often necessary to monitor cracks and damage in pipelines, or need to clean the inside of the pipeline regularly, or collect adhesive on the inner wall of the pipe, but the pipe is too narrow and difficult for humans to enter, it is necessary to use a pipe machine to complete the work. In this paper, a newly designed screw-driven in-pipe inspection robot (IPIR) is proposed. Compared with common robots, this robot innovatively designs adapting mechanism. The robot can not only adapt to the change of the inner diameter size of the pipeline by using the bionic principle and the deformation characteristics of flexible components but also can pass smoothly in the horizontal/oblique/vertical pipelines and has a certain ability to cross obstacles. In addition, it can transmit images of the inner wall of the pipeline wirelessly for data analysis. Finally, through theoretical analysis and prototype construction, the performance of the robot is verified. The results show that the prototype robot can not only smoothly pass through the acrylic pipe with inner diameter of 120–138 mm but also pass through boss with a height of 3 mm.