Pub Date : 2025-09-24DOI: 10.1126/scirobotics.aeb6685
Robin R. Murphy
The novel Flybot gives a science-forward view of the challenges in building a fully autonomous robot fly.
新颖的Flybot给出了一个科学的观点，在建立一个完全自主的机器人飞行的挑战。
{"title":"Lowly fly or impressive miniature robot?","authors":"Robin R. Murphy","doi":"10.1126/scirobotics.aeb6685","DOIUrl":"10.1126/scirobotics.aeb6685","url":null,"abstract":"<div >The novel <i>Flybot</i> gives a science-forward view of the challenges in building a fully autonomous robot fly.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145133973","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-17DOI: 10.1126/scirobotics.adu4851
Xurui Liu, Hanchuan Tang, Na Li, Linjie He, Ye Tian, Bo Hao, Junnan Xue, Chaoyu Yang, Joseph Jao Yiu Sung, Li Zhang, Jianfeng Zang
Intelligent miniature systems capable of wireless sensing and manipulation hold considerable promise for advancing biomedical applications. However, the development of these systems has been substantially hindered by sensing-actuation incompatibility at small scales. To overcome this challenge, we propose a robotic sensing approach that integrates embedded ultrasonic soft sensors (EUSSs) with magnetic actuators, resulting in a wireless sensor-integrated miniature machine with seamless integration and minimal interference between fields. The EUSS, with its compact dimensions (1.3 millimeters by 1.3 millimeters by 1.6 millimeters), softness (98 kilopascals), and lightweight design (4.6 milligrams), is compatible with both soft and rigid components in terms of deformability and size. By engineering onboard transducers and using passive ultrasound communication along with external magnetic fields, we could wirelessly detect and regulate environmental parameters such as force, vibration, viscosity, and temperature. Demonstrations in rabbit and porcine models show the potential for robotic feedback control, accurate drug dosing, and in situ physiological monitoring, paving the way for real-world applications of intelligent miniature machines.
{"title":"Miniature magneto-ultrasonic machines for wireless robotic sensing and manipulation","authors":"Xurui Liu, Hanchuan Tang, Na Li, Linjie He, Ye Tian, Bo Hao, Junnan Xue, Chaoyu Yang, Joseph Jao Yiu Sung, Li Zhang, Jianfeng Zang","doi":"10.1126/scirobotics.adu4851","DOIUrl":"10.1126/scirobotics.adu4851","url":null,"abstract":"<div >Intelligent miniature systems capable of wireless sensing and manipulation hold considerable promise for advancing biomedical applications. However, the development of these systems has been substantially hindered by sensing-actuation incompatibility at small scales. To overcome this challenge, we propose a robotic sensing approach that integrates embedded ultrasonic soft sensors (EUSSs) with magnetic actuators, resulting in a wireless sensor-integrated miniature machine with seamless integration and minimal interference between fields. The EUSS, with its compact dimensions (1.3 millimeters by 1.3 millimeters by 1.6 millimeters), softness (98 kilopascals), and lightweight design (4.6 milligrams), is compatible with both soft and rigid components in terms of deformability and size. By engineering onboard transducers and using passive ultrasound communication along with external magnetic fields, we could wirelessly detect and regulate environmental parameters such as force, vibration, viscosity, and temperature. Demonstrations in rabbit and porcine models show the potential for robotic feedback control, accurate drug dosing, and in situ physiological monitoring, paving the way for real-world applications of intelligent miniature machines.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145077510","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-17DOI: 10.1126/scirobotics.aea9377
Xuhui Hu, Aiguo Song, Min Xu
Innovations in sensing and control technology helped an arm prosthesis novice win a global assistive robotics competition.
传感和控制技术的创新帮助手臂假肢新手赢得全球辅助机器人比赛。
{"title":"Arm prosthesis with dexterous control and sensory feedback delivers winning performance at Cybathlon","authors":"Xuhui Hu, Aiguo Song, Min Xu","doi":"10.1126/scirobotics.aea9377","DOIUrl":"10.1126/scirobotics.aea9377","url":null,"abstract":"<div >Innovations in sensing and control technology helped an arm prosthesis novice win a global assistive robotics competition.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.aea9377","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145078231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-17DOI: 10.1126/scirobotics.aeb1340
Garrett Kryt, Rory Dougall, Jaimie Borisoff
An extending, articulating powered wheelchair competed and won the wheelchair race at Cybathlon 2024.
在2024年Cybathlon的轮椅比赛中，一款可伸缩的动力轮椅赢得了冠军。
{"title":"BCIT’s BEAST wheelchair takes on Cybathlon with power, precision, and pilot-led design","authors":"Garrett Kryt, Rory Dougall, Jaimie Borisoff","doi":"10.1126/scirobotics.aeb1340","DOIUrl":"10.1126/scirobotics.aeb1340","url":null,"abstract":"<div >An extending, articulating powered wheelchair competed and won the wheelchair race at Cybathlon 2024.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.aeb1340","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145078357","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-17DOI: 10.1126/scirobotics.adt1497
Bo Ai, Stephen Tian, Haochen Shi, Yixuan Wang, Tobias Pfaff, Cheston Tan, Henrik I. Christensen, Hao Su, Jiajun Wu, Yunzhu Li
Dynamics models that predict the effects of physical interactions are essential for planning and control in robotic manipulation. Although models based on physical principles often generalize well, they typically require full-state information, which can be difficult or impossible to extract from perception data in complex, real-world scenarios. Learning-based dynamics models provide an alternative by deriving state transition functions purely from perceived interaction data, enabling the capture of complex, hard-to-model factors and predictive uncertainty and accelerating simulations that are often too slow for real-time control. Recent successes in this field have demonstrated notable advancements in robot capabilities, including long-horizon manipulation of deformable objects, granular materials, and complex multiobject interactions such as stowing and packing. A crucial aspect of these investigations is the choice of state representation, which determines the inductive biases in the learning system for reduced-order modeling of scene dynamics. This article provides a timely and comprehensive review of current techniques and trade-offs in designing learned dynamics models, highlighting their role in advancing robot capabilities through integration with state estimation and control and identifying critical research gaps for future exploration.
{"title":"A review of learning-based dynamics models for robotic manipulation","authors":"Bo Ai, Stephen Tian, Haochen Shi, Yixuan Wang, Tobias Pfaff, Cheston Tan, Henrik I. Christensen, Hao Su, Jiajun Wu, Yunzhu Li","doi":"10.1126/scirobotics.adt1497","DOIUrl":"10.1126/scirobotics.adt1497","url":null,"abstract":"<div >Dynamics models that predict the effects of physical interactions are essential for planning and control in robotic manipulation. Although models based on physical principles often generalize well, they typically require full-state information, which can be difficult or impossible to extract from perception data in complex, real-world scenarios. Learning-based dynamics models provide an alternative by deriving state transition functions purely from perceived interaction data, enabling the capture of complex, hard-to-model factors and predictive uncertainty and accelerating simulations that are often too slow for real-time control. Recent successes in this field have demonstrated notable advancements in robot capabilities, including long-horizon manipulation of deformable objects, granular materials, and complex multiobject interactions such as stowing and packing. A crucial aspect of these investigations is the choice of state representation, which determines the inductive biases in the learning system for reduced-order modeling of scene dynamics. This article provides a timely and comprehensive review of current techniques and trade-offs in designing learned dynamics models, highlighting their role in advancing robot capabilities through integration with state estimation and control and identifying critical research gaps for future exploration.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145077693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-10DOI: 10.1126/scirobotics.adu5771
Lauren L. Wright, Pooja Vegesna, Joseph E. Michaelis, Bilge Mutlu, Sarah Sebo
Reading fluency is a vital building block for developing literacy, yet the best way to practice fluency—reading aloud—can cause anxiety severe enough to inhibit literacy development in ways that can have an adverse effect on students through adulthood. One promising intervention to mitigate oral reading anxiety is to have children read aloud to a robot. Although observations in prior work have suggested that people likely feel more comfortable in the presence of a robot instead of a human, few studies have empirically demonstrated that people feel less anxious performing in front of a robot compared with a human or used objective physiological indicators to identify decreased anxiety. To investigate whether a robotic reading companion could reduce reading anxiety felt by children, we conducted a within-subjects study where children aged 8 to 11 years (n = 52) read aloud to a human and a robot individually while being monitored for physiological responses associated with anxiety. We found that children exhibited fewer physiological indicators of anxiety, specifically vocal jitter and heart rate variability, when reading to the robot compared with reading to a person. This paper provides strong evidence that a robot’s presence has an effect on the anxiety a person experiences while doing a task, offering justification for the use of robots in a wide-reaching array of social interactions that may be anxiety inducing.
{"title":"Robotic reading companions can mitigate oral reading anxiety in children","authors":"Lauren L. Wright, Pooja Vegesna, Joseph E. Michaelis, Bilge Mutlu, Sarah Sebo","doi":"10.1126/scirobotics.adu5771","DOIUrl":"10.1126/scirobotics.adu5771","url":null,"abstract":"<div >Reading fluency is a vital building block for developing literacy, yet the best way to practice fluency—reading aloud—can cause anxiety severe enough to inhibit literacy development in ways that can have an adverse effect on students through adulthood. One promising intervention to mitigate oral reading anxiety is to have children read aloud to a robot. Although observations in prior work have suggested that people likely feel more comfortable in the presence of a robot instead of a human, few studies have empirically demonstrated that people feel less anxious performing in front of a robot compared with a human or used objective physiological indicators to identify decreased anxiety. To investigate whether a robotic reading companion could reduce reading anxiety felt by children, we conducted a within-subjects study where children aged 8 to 11 years (<i>n</i> = 52) read aloud to a human and a robot individually while being monitored for physiological responses associated with anxiety. We found that children exhibited fewer physiological indicators of anxiety, specifically vocal jitter and heart rate variability, when reading to the robot compared with reading to a person. This paper provides strong evidence that a robot’s presence has an effect on the anxiety a person experiences while doing a task, offering justification for the use of robots in a wide-reaching array of social interactions that may be anxiety inducing.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145028467","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-10DOI: 10.1126/scirobotics.adu6123
Joseph E. Michaelis, Bilge Mutlu
Family-centered integration is critical for the success of in-home educational robots.
以家庭为中心的整合对于家庭教育机器人的成功至关重要。
{"title":"How can educational robots enhance family life? Through careful integration","authors":"Joseph E. Michaelis, Bilge Mutlu","doi":"10.1126/scirobotics.adu6123","DOIUrl":"10.1126/scirobotics.adu6123","url":null,"abstract":"<div >Family-centered integration is critical for the success of in-home educational robots.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145028495","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
According to productive failure (PF) theory, experiencing failure during problem-solving can enhance students’ knowledge acquisition in subsequent instruction. However, challenging students with problems beyond their current capabilities may strain their skills, prior knowledge, and emotional well-being. To address this, we designed a social robot–assisted teaching activity in which students observed a robot’s unsuccessful problem-solving attempts, offering a PF-like preparatory effect without requiring direct failure. We conducted two classroom-based studies in a middle school setting to evaluate the method’s effectiveness. In study 1 (N = 135), we compared three instructional methods—observing robot failure (RF), individual problem-solving failure, and direct instruction—in an eighth-grade mathematics lesson. Students in the RF condition showed the greatest gains in conceptual understanding and reported lower social pressure, although no significant differences were found in procedural knowledge or knowledge transfer. Follow-up study 2 (N = 110) further validated the method’s effectiveness in supporting knowledge acquisition after a 2-week robot-involved adaptation phase, when the novelty effect had largely subsided. Students confirmed their perception of the robot as a peer, and they offered positive evaluations of its intelligence and neutral views of its anthropomorphism. Our findings suggest that observing the robot’s failure has a comparable, or even greater, effect on knowledge acquisition than experiencing failure firsthand. These results underscore the value of social robots as peers in science, technology, engineering, and mathematics education and highlight the potential of integrating robotics with evidence-based teaching strategies to enhance learning outcomes.
{"title":"Observing a robot peer’s failures facilitates students’ classroom learning","authors":"Liuqing Chen, Yu Cai, Yuyang Fang, Ziqi Yang, Duowei Xia, Jiaxiang You, Shuhong Xiao, Yaxuan Song, Lingwei Zhan, Juanjuan Chen, Lingyun Sun","doi":"10.1126/scirobotics.adu5257","DOIUrl":"10.1126/scirobotics.adu5257","url":null,"abstract":"<div >According to productive failure (PF) theory, experiencing failure during problem-solving can enhance students’ knowledge acquisition in subsequent instruction. However, challenging students with problems beyond their current capabilities may strain their skills, prior knowledge, and emotional well-being. To address this, we designed a social robot–assisted teaching activity in which students observed a robot’s unsuccessful problem-solving attempts, offering a PF-like preparatory effect without requiring direct failure. We conducted two classroom-based studies in a middle school setting to evaluate the method’s effectiveness. In study 1 (<i>N</i> = 135), we compared three instructional methods—observing robot failure (RF), individual problem-solving failure, and direct instruction—in an eighth-grade mathematics lesson. Students in the RF condition showed the greatest gains in conceptual understanding and reported lower social pressure, although no significant differences were found in procedural knowledge or knowledge transfer. Follow-up study 2 (<i>N</i> = 110) further validated the method’s effectiveness in supporting knowledge acquisition after a 2-week robot-involved adaptation phase, when the novelty effect had largely subsided. Students confirmed their perception of the robot as a peer, and they offered positive evaluations of its intelligence and neutral views of its anthropomorphism. Our findings suggest that observing the robot’s failure has a comparable, or even greater, effect on knowledge acquisition than experiencing failure firsthand. These results underscore the value of social robots as peers in science, technology, engineering, and mathematics education and highlight the potential of integrating robotics with evidence-based teaching strategies to enhance learning outcomes.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145028483","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-03DOI: 10.1126/scirobotics.adu5830
Hyegi Min, Yue Wang, Jiaojiao Wang, Xiuyuan Li, Woong Kim, Onur Aydin, Sehong Kang, Jae-Sung You, Jongwon Lim, Katy Wolhaupter, Yikang Xu, Zhengguang Zhu, Jianyu Gu, Xinming Li, Yongdeok Kim, Tarun Rao, Hyun Joon Kong, Taher A. Saif, Yonggang Huang, John A. Rogers, Rashid Bashir
Neuronal control of skeletal muscle function is ubiquitous across species for locomotion and doing work. In particular, emergent behaviors of neurons in biohybrid neuromuscular systems can advance bioinspired locomotion research. Although recent studies have demonstrated that chemical or optogenetic stimulation of neurons can control muscular actuation through the neuromuscular junction (NMJ), the correlation between neuronal activities and resulting modulation in the muscle responses is less understood, hindering the engineering of high-level functional biohybrid systems. Here, we developed NMJ-based biohybrid crawling robots with optogenetic mouse motor neurons, skeletal muscles, 3D-printed hydrogel scaffolds, and integrated onboard wireless micro–light-emitting diode (μLED)–based optoelectronics. We investigated the coupling of the light stimulation and neuromuscular actuation through power spectral density (PSD) analysis. We verified the modulation of the mechanical functionality of the robot depending on the frequency of the optical stimulation to the neural tissue. We demonstrated continued muscle contraction up to 20 minutes after a 1-minute-long pulsed 2-hertz optical stimulation of the neural tissue. Furthermore, the robots were shown to maintain their mechanical functionality for more than 2 weeks. This study provides insights into reliable neuronal control with optoelectronics, supporting advancements in neuronal modulation, biohybrid intelligence, and automation.
{"title":"Optogenetic neuromuscular actuation of a miniature electronic biohybrid robot","authors":"Hyegi Min, Yue Wang, Jiaojiao Wang, Xiuyuan Li, Woong Kim, Onur Aydin, Sehong Kang, Jae-Sung You, Jongwon Lim, Katy Wolhaupter, Yikang Xu, Zhengguang Zhu, Jianyu Gu, Xinming Li, Yongdeok Kim, Tarun Rao, Hyun Joon Kong, Taher A. Saif, Yonggang Huang, John A. Rogers, Rashid Bashir","doi":"10.1126/scirobotics.adu5830","DOIUrl":"10.1126/scirobotics.adu5830","url":null,"abstract":"<div >Neuronal control of skeletal muscle function is ubiquitous across species for locomotion and doing work. In particular, emergent behaviors of neurons in biohybrid neuromuscular systems can advance bioinspired locomotion research. Although recent studies have demonstrated that chemical or optogenetic stimulation of neurons can control muscular actuation through the neuromuscular junction (NMJ), the correlation between neuronal activities and resulting modulation in the muscle responses is less understood, hindering the engineering of high-level functional biohybrid systems. Here, we developed NMJ-based biohybrid crawling robots with optogenetic mouse motor neurons, skeletal muscles, 3D-printed hydrogel scaffolds, and integrated onboard wireless micro–light-emitting diode (μLED)–based optoelectronics. We investigated the coupling of the light stimulation and neuromuscular actuation through power spectral density (PSD) analysis. We verified the modulation of the mechanical functionality of the robot depending on the frequency of the optical stimulation to the neural tissue. We demonstrated continued muscle contraction up to 20 minutes after a 1-minute-long pulsed 2-hertz optical stimulation of the neural tissue. Furthermore, the robots were shown to maintain their mechanical functionality for more than 2 weeks. This study provides insights into reliable neuronal control with optoelectronics, supporting advancements in neuronal modulation, biohybrid intelligence, and automation.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144935569","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-03DOI: 10.1126/scirobotics.ads1204
Matthew Lai, Keegan Go, Zhibin Li, Torsten Kröger, Stefan Schaal, Kelsey Allen, Jonathan Scholz
Modern robotic manufacturing requires collision-free coordination of multiple robots to complete numerous tasks in shared, obstacle-rich workspaces. Although individual tasks may be simple in isolation, automated joint task allocation, scheduling, and motion planning under spatiotemporal constraints remain computationally intractable for classical methods at real-world scales. Existing multiarm systems deployed in industry rely on human intuition and experience to design feasible trajectories manually in a labor-intensive process. To address this challenge, we propose a reinforcement learning (RL) framework to achieve automated task and motion planning, tested in an obstacle-rich environment with eight robots performing 40 reaching tasks in a shared workspace, where any robot can perform any task in any order. Our approach builds on a graph neural network (GNN) policy trained via RL on procedurally generated environments with diverse obstacle layouts, robot configurations, and task distributions. It uses a graph representation of scenes and a graph policy neural network trained through RL to generate trajectories of multiple robots, jointly solving the subproblems of task allocation, scheduling, and motion planning. Trained on large randomly generated task sets in simulation, our policy generalizes zero-shot to unseen settings with varying robot placements, obstacle geometries, and task poses. We further demonstrate that the high-speed capability of our solution enables its use in workcell layout optimization, improving solution times. The speed and scalability of our planner also open the door to capabilities such as fault-tolerant planning and online perception-based replanning, where rapid adaptation to dynamic task sets is required.
{"title":"RoboBallet: Planning for multirobot reaching with graph neural networks and reinforcement learning","authors":"Matthew Lai, Keegan Go, Zhibin Li, Torsten Kröger, Stefan Schaal, Kelsey Allen, Jonathan Scholz","doi":"10.1126/scirobotics.ads1204","DOIUrl":"10.1126/scirobotics.ads1204","url":null,"abstract":"<div >Modern robotic manufacturing requires collision-free coordination of multiple robots to complete numerous tasks in shared, obstacle-rich workspaces. Although individual tasks may be simple in isolation, automated joint task allocation, scheduling, and motion planning under spatiotemporal constraints remain computationally intractable for classical methods at real-world scales. Existing multiarm systems deployed in industry rely on human intuition and experience to design feasible trajectories manually in a labor-intensive process. To address this challenge, we propose a reinforcement learning (RL) framework to achieve automated task and motion planning, tested in an obstacle-rich environment with eight robots performing 40 reaching tasks in a shared workspace, where any robot can perform any task in any order. Our approach builds on a graph neural network (GNN) policy trained via RL on procedurally generated environments with diverse obstacle layouts, robot configurations, and task distributions. It uses a graph representation of scenes and a graph policy neural network trained through RL to generate trajectories of multiple robots, jointly solving the subproblems of task allocation, scheduling, and motion planning. Trained on large randomly generated task sets in simulation, our policy generalizes zero-shot to unseen settings with varying robot placements, obstacle geometries, and task poses. We further demonstrate that the high-speed capability of our solution enables its use in workcell layout optimization, improving solution times. The speed and scalability of our planner also open the door to capabilities such as fault-tolerant planning and online perception-based replanning, where rapid adaptation to dynamic task sets is required.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"10 106","pages":""},"PeriodicalIF":27.5,"publicationDate":"2025-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144935547","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：