Robotics and Autonomous Systems最新文献

{"title":"Deep reinforcement learning-based design with observation buffer of rolling motion for snake-like robots","authors":"Akio Yamano,&nbsp;Satomi Suzuki,&nbsp;Tsuyoshi Kimoto,&nbsp;Takashi Iwasa","doi":"10.1016/j.robot.2026.105370","DOIUrl":"10.1016/j.robot.2026.105370","url":null,"abstract":"<div><div>Snake-like robots can travel over environments that are difficult for wheeled mobility mechanisms. However, undulating locomotion requires high power consumption. We propose an efficient method that integrates the center-of-gravity (COG) shifting for the navigation of the robot to address the aforementioned problem. In the proposed method, the snake-like robot transforms into a tire-like shape to realize a parallel two-wheeled configuration. Subsequently, by deforming the head or tail sections, the position of the COG is changed, and the resulting gravitational torque generates a rolling motion. The proposed method allows the use of rolling motion with high traveling efficiency on level ground and undulating locomotion in water as well as other uneven surfaces. The rolling motion designed in previous research was achieved by the feedback of the direction of gravity measured by an acceleration sensor. Therefore, it was only designed to be capable of traveling on smooth floors or asphalt, making it difficult to maintain straight traveling when road conditions change. This paper presents a controller design method using deep reinforcement learning (RL) to achieve robust traveling by the rolling motion. We conducted experiments using the controller designed by RL and compared the experimental results with numerical simulations. Experiments demonstrated that the RL-designed rolling motion achieved higher straightness than that of previous methods and higher traveling efficiency than conventional undulating locomotion.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105370"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174551","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrated trajectory planning and tracking control for FWID-AV collision avoidance with multi-actuator control distribution","authors":"Zezhao Wang ,&nbsp;Conggan Ma ,&nbsp;Yi Li ,&nbsp;Ruiyi Wu","doi":"10.1016/j.robot.2026.105355","DOIUrl":"10.1016/j.robot.2026.105355","url":null,"abstract":"<div><div>As a major development direction of modern automobiles, autonomous vehicles can significantly improve driving safety, efficiency, and economic benefits, making them an urgent societal need. However, existing reports often overlook the impact of prediction inaccuracies in obstacle motion, particularly in complex traffic scenarios. In addition, local trajectory planning and tracking are typically treated as separate tasks, which may lead to poor tracking accuracy of the local reference trajectory and even controller failure. With that in mind, this paper focuses on four-wheel-independent-drive autonomous vehicles (FWID-AV) and proposes an integrated control framework that considers local trajectory planning and trajectory tracking for dynamic obstacle avoidance based on Model Predictive Control (MPC) and Particle Swarm Optimization (PSO) algorithms. Firstly, the vehicle dynamics model is established considering the specific functional requirements of AV trajectory planning and tracking control. Then, a new planning-integrated controller is developed for FWID-AV to achieve collision avoidance. Finally, the MATLAB/Simulink-CarSim joint simulation platform is built, and the effectiveness of the proposed integrated control framework is verified through simulation and experimental tests, which enhances the performance of collision avoidance, trajectory tracking, handling stability, and real-time performance for vehicle.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105355"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A half-cooperative strategy based Deep Q-network algorithm for multi-agent dynamic target hunting","authors":"Xiaoyan Wang,&nbsp;Xin Yu,&nbsp;Xi Fang","doi":"10.1016/j.robot.2026.105368","DOIUrl":"10.1016/j.robot.2026.105368","url":null,"abstract":"<div><div>In fields like rescue, military, and law enforcement, effective target hunting is vital. Current algorithms for dynamic target hunting often fail in multi-target scenarios due to unknown environments and low cooperation efficiency. This paper introduces a multi-agent cooperative task planning model with a novel half-cooperative DQN algorithm for dynamic target hunting in unknown environments. The model randomly assigns positions to agents and targets, defines the action space for agents, and designs reward functions for real-world situations. The half-cooperative DQN algorithm uses prioritized experience replay for efficient learning and employs a half-cooperative strategy to enhance cooperation among agents, thereby improving hunting efficiency. Experimental results show that the half-cooperative DQN algorithm outperforms other improved DQN algorithms in terms of success rate, average boundary violations, and average time steps, highlighting its advantages and potential in dynamic target hunting.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105368"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146070764","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CMAE-Traj: A contrastive masked autoencoder framework for trajectory prediction","authors":"Yamei Xu ,&nbsp;Hong Zhang ,&nbsp;Zhenghan Gao ,&nbsp;Chengming Liu ,&nbsp;Zhe Zhao","doi":"10.1016/j.robot.2026.105391","DOIUrl":"10.1016/j.robot.2026.105391","url":null,"abstract":"<div><div>Predicting the future trajectories of surrounding agents is crucial for the safe operation of autonomous vehicles. Self-supervised learning has gained widespread attention in trajectory prediction, as it alleviates the high cost of annotating trajectory data by pretraining on unlabeled data. However, due to limitations in capturing spatio-temporal interactions and decoupling short-term dynamics and long-term trends, current self-supervised learning methods struggle to model complex behaviors between agents. To overcome these challenges, we propose a novel masked autoencoder with contrastive learning for trajectory prediction that effectively extracts the complicated interactions of agents in the driving environment. Specifically, we introduce a temporal-social mask module that captures short-term motion patterns by jointly modeling temporal dynamics and social interactions of agents, enabling the model to gain a comprehensive understanding from multiple perspectives. Moreover, we introduce a contrastive mask alignment strategy that learns consistent long-term motion trends by treating features from the same trajectory as positive samples and features from different trajectories as negative samples. Extensive experiments on the Argoverse 2, INTERACTION and Waymo Open Motion datasets demonstrate that our model significantly outperforms previous self-supervised learning approaches, achieving competitive results in trajectory prediction.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105391"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174897","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SafeIL: Safety constrained imitation learning for autonomous systems","authors":"Gunmin Lee ,&nbsp;Jaeseok Heo ,&nbsp;Dohyeong Kim ,&nbsp;Geunje Cheon ,&nbsp;Jeongwoo Oh ,&nbsp;Minyoung Hwang ,&nbsp;Chanwoo Park ,&nbsp;Kyungjae Lee ,&nbsp;Songhwai Oh","doi":"10.1016/j.robot.2026.105376","DOIUrl":"10.1016/j.robot.2026.105376","url":null,"abstract":"<div><div>Safety is a critical concern in controller design, yet developing a cost function that accurately reflects safety remains a significant challenge, akin to the complexities of designing a reward function. To address this, we introduce Safety Constrained Imitation Learning (SafeIL), an innovative safety-constrained imitation learning framework that simultaneously estimates reward and safety cost functions using two distinct sets of expert demonstrations: one aimed at maximizing rewards without considering safety, and the other focused on avoiding safety violations during execution. Through the use of dual independent discriminator networks, SafeIL effectively learns these functions, enabling the development of a controller that ensures safety while maintaining high performance. Our empirical evaluations across diverse simulation environments, including safety-gym, Metadrive, CARLA, F1tenth, and real-world platforms such as Jackal and RC car, demonstrate that SafeIL significantly outperforms existing methods like GAIL. Specifically, SafeIL achieves substantial reductions in constraint violations, including zero violations on the Jackal platform, and reduces violations by 79.6% compared to GAIL when utilizing safety-focused demonstrations, underscoring its potential to enhance safety in real-world robotic applications.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105376"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174555","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"VLM-integrated 3D perception model for robust robotic grasping adapted to deformable sacks with arbitrary shapes","authors":"Seongje Kim ,&nbsp;Jonghun Yoon","doi":"10.1016/j.robot.2026.105372","DOIUrl":"10.1016/j.robot.2026.105372","url":null,"abstract":"<div><div>This study presents a vision-language and 3D perception-integrated system for robust grasping of non-standard logistics parcels, including untrained boxes and deformable sacks. To address the limitations of conventional recognition models in handling unseen or irregularly shaped objects, we propose a zero-shot object recognition framework based on a Vision-Language Model (VLM), combined with confidence threshold optimization and Non-Maximum Suppression (NMS) to improve detection reliability. The system further incorporates 3D point cloud-based post-processing to extract precise grasp points adapted to both rigid and deformable object geometries. Proposed approach achieves a mean Average Precision (mAP) of 80.1 % for detecting untrained boxes and burlap sacks and demonstrates a grasp success rate of 99.0 % for boxes and 73.0 % for deformable sacks in real-world unloading scenarios using collaborative robots. Unlike prior methods that require extensive retraining for new object types, proposed system enables generalizable, real-time grasping without additional dataset preparation. The hybrid integration of rule-based and learning-based strategies in 3D space contributes to sTable and adapTable grasp point selection across variable object types. This work substantiates the feasibility of zero-shot recognition and grasping of non-standard logistics items, offering a practical solution for automated parcel unloading in dynamic industrial environments.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105372"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174556","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Collision-free multi-UAV 3D path planning using dynamic spatial reservations of reducing air corridors","authors":"Andrey Ronzhin ,&nbsp;Anton Saveliev ,&nbsp;Dmitry Anikin ,&nbsp;Alexandra Zaytseva ,&nbsp;Ekaterina Cherskikh ,&nbsp;Aleksandra Figurek","doi":"10.1016/j.robot.2026.105359","DOIUrl":"10.1016/j.robot.2026.105359","url":null,"abstract":"<div><div>This paper presents a centralized coordination framework for multi-agent Unmanned Aerial Vehicle (UAV) systems, designed to ensure collision-free navigation in complex three-dimensional environments. The core contribution is a dynamic spatial reservation mechanism based on reducing air corridors. These defined, capsule-shaped volumes are allocated to each agent and contract along their longitudinal axis during traversal, thereby progressively releasing airspace for subsequent vehicles. To enhance safety during coordination failures or at predefined target points, agents are encapsulated within protective spherical zones. The coordination logic adheres to a first-come-first-served sequential planning paradigm, wherein agents employ an any-angle path planner. Generated paths are constrained to avoid both static obstacles and reserved corridors, while adhering to the agent's specific safe-radius constraints. In the event of a planning failure induced by spatial conflicts, agents enter a protective state and re-attempt path generation after a configurable delay. The collision detection subsystem implements a multi-layered approach for efficiency. Broad-phase filtering utilizes an octree for static obstacles, a hierarchical hash grid for dynamic corridors, and elevation bounds for terrain. The narrow phase employs analytical geometric verification, modeling static obstacles as convex polyhedra, dynamic corridors as capsules, and terrain as bilinear patches derived from elevation grids. This ensures precise, exact-distance computations for all collision checks. Experimental results validate the framework's efficacy in achieving collision-free coordination across a range of operational scenarios. Although computational load and agent waiting times exhibit a predictable increase with population density, the method provides formal guarantees of safety and mission completion through its rigorous validation and retry mechanisms. Consequently, the proposed approach is particularly suited for safety-critical applications where operational reliability is paramount.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105359"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174554","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Offshore Subsea IMR Operations: Review of the Automation Potential","authors":"Jannic Schurmann Larsen,&nbsp;Simon Pedersen,&nbsp;Jesper Liniger,&nbsp;Fredrik Fogh Sørensen","doi":"10.1016/j.robot.2026.105365","DOIUrl":"10.1016/j.robot.2026.105365","url":null,"abstract":"<div><div>The demand for green energy continues to rise, with predictions indicating a 62-fold growth in floating wind. This offshore expansion, especially from wind, requires a corresponding increase in inspection, maintenance, and repair (IMR) operations. For offshore turbines, the marine segment accounts for half of fixed turbine O&amp;M costs. One avenue to reduce these costs is through unmanned underwater vehicles (UUV) and automation. UUVs are becoming more specialized and automated, but most tasks remain only partially automated. It is estimated that the global energy cost in 2050 can be reduced by 0.5–0.6% from increased unmanned underwater vehicle automation, with wind contributing 0.4–0.5%. This study assesses automation levels and task frequencies across application domains, defined as types of offshore infrastructure, to calculate an automation priority score (APS), a weighted product of automation level and task frequency, indicating where economic and development benefits are. The automation level refers to a UUVs ability to operate without human intervention while performing tasks, regardless of industry use. The current highest APS is found in pipelines and jacket structures, with the lowest in wind turbines and dams. A sensitivity analysis evaluates the effects of increasing automation levels and anticipated annual growth. The study concludes that the wind sector will represent the highest automation priority in the future, providing the greatest economic incentive, especially for mooring lines. Visual inspection will increase due to AI, crawler-based UUVs will dominate circular structures IMR tasks, and autonomous underwater vehicle (AUVs) with subsea stations will handle frequent or long-range tasks.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"199 ","pages":"Article 105365"},"PeriodicalIF":5.2,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146070765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Terrain-based place recognition for LiDAR SLAM of quadruped robots with limited field-of-view measurements","authors":"Roun Lee ,&nbsp;Seonghun Hong ,&nbsp;Sukmin Yoon","doi":"10.1016/j.robot.2025.105315","DOIUrl":"10.1016/j.robot.2025.105315","url":null,"abstract":"<div><div>Over the past few decades, light detection and ranging (LiDAR) sensors have been extensively employed for pose estimation in simultaneous localization and mapping (SLAM). In more recent years, the use of solid-state LiDAR sensors with no rotating mechanisms and a limited field-of-view for SLAM has attracted research attention because of their cost effectiveness and durability. However, it is highly challenging to successfully perform place recognition, which is one of the most important components of SLAM, via limited field-of-view measurements. Failure in place recognition can severely degrade the resulting estimation performance of SLAM algorithms. Considering a terrestrial SLAM framework for quadruped robots with limited field-of-view LiDAR sensors, this study proposes a terrain-based place recognition algorithm that reconstructs and compares detected feature terrains, using a set of foot contact information for quadruped robots. The validity and practical feasibility of the proposed approach are demonstrated through experimental results using a quadruped robot system with a limited field-of-view LiDAR sensor.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105315"},"PeriodicalIF":5.2,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145939734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploiting Euclidean distance field properties for fast and safe 3D planning with a modified Lazy Theta*","authors":"Jose A. Cobano,&nbsp;L. Merino,&nbsp;F. Caballero","doi":"10.1016/j.robot.2025.105317","DOIUrl":"10.1016/j.robot.2025.105317","url":null,"abstract":"<div><div>This paper presents the FS-Planner, a fast graph-search planner based on a modified Lazy Theta* algorithm that exploits the analytical properties of Euclidean Distance Fields (EDFs). We introduce a new cost function that integrates an EDF-based term proven to satisfy the triangle inequality, enabling efficient parent selection and reducing computation time while generating safe paths with smaller heading variations. We also derive an analytic approximation of the EDF integral along a segment and analyse the influence of the line-of-sight limit on the approximation error, motivating the use of a bounded visibility range. Furthermore, we propose a gradient-based neighbour-selection mechanism that decreases the number of explored nodes and improves computational performance without degrading safety or path quality. The FS-Planner produces safe paths with small heading changes without requiring the use of post-processing methods. Extensive experiments and comparisons in challenging 3D indoor simulation environments, complemented by tests in real-world outdoor environments, are used to evaluate and validate the FS-Planner. The results show consistent improvements in computation time, exploration efficiency, safety, and smoothness in a geometric sense compared with baseline heuristic planners, while maintaining sub-optimality within acceptable bounds. Finally, the proposed EDF-based cost formulation is orthogonal to the underlying search method and can be incorporated into other planning paradigms.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105317"},"PeriodicalIF":5.2,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145940044","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}