Science Robotics最新文献

Structural bias enables reliable sim-to-real transfer in reinforcement learning for robot swarms 结构偏差在机器人群体强化学习中实现了可靠的模拟到真实的迁移

IF 25 1区计算机科学

Science Robotics Pub Date : 2026-10-21 DOI: 10.1126/scirobotics.adw4624

Ilyes Gharbi, Miquel Kegeleirs, Guillermo Legarda Herranz, Jeanne Szpirer, Ken Hasselmann, Mauro Birattari

{"title":"Structural bias enables reliable sim-to-real transfer in reinforcement learning for robot swarms","authors":"Ilyes Gharbi, Miquel Kegeleirs, Guillermo Legarda Herranz, Jeanne Szpirer, Ken Hasselmann, Mauro Birattari","doi":"10.1126/scirobotics.adw4624","DOIUrl":"https://doi.org/10.1126/scirobotics.adw4624","url":null,"abstract":"Reinforcement learning offers a powerful framework for automatically generating control policies for robot swarms, yet its application to physical multirobot systems has been limited by the simulation-to-reality gap. Despite substantial advances in multiagent reinforcement learning, systematic demonstrations of gradient-based sim-to-real transfer to physical robot swarms remain scarce. Most work remains confined to simulation, and the few real-robot demonstrations lack evaluation across multiple runs and tasks. We show that this barrier can be overcome by embedding structural constraints into policy learning. We introduce attention-based counterfactual policy gradients for behavior selection in robot swarms (SwarmACB), a multiagent actor-critic method in which robots arbitrate among predefined behavior modules—reactive sensorimotor routines such as phototaxis or aggregation—rather than learning low-level motor commands. This architectural choice reduces the variance of learned solutions, which we argue mitigates overfitting to simulation artifacts. Through systematic evaluation on five collective missions using 20 physical e-puck robots (250 real-robot runs), we demonstrate that SwarmACB transfers successfully from simulation to reality, substantially outperforming continuous-action baselines. By progressively constraining policy architecture, perception, and model capacity, we show that smaller, structured policies transfer more reliably than large, unconstrained ones. These findings challenge the assumption that scale is necessary for successful transfer, suggesting instead that principled structural design enables reinforcement learning to produce control software for robot swarms that is both learned automatically and directly deployable in the physical world.","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"35 1","pages":""},"PeriodicalIF":25.0,"publicationDate":"2026-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149226144","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}

引用次数: 0

Hearing the shape of an arena with spectral swarm robotics 用光谱蜂群机器人听到竞技场的形状

IF 25 1区计算机科学

Science Robotics Pub Date : 2026-10-21 DOI: 10.1126/scirobotics.adr9578

Leo Cazenille, Nicolas Lobato-Dauzier, Alessia Loi, Loona Macabre, Mika Ito, Olivier Marchal, Soo-Hyeon Kim, Nathanael Aubert-Kato, Nicolas Bredeche, Anthony J. Genot

{"title":"Hearing the shape of an arena with spectral swarm robotics","authors":"Leo Cazenille, Nicolas Lobato-Dauzier, Alessia Loi, Loona Macabre, Mika Ito, Olivier Marchal, Soo-Hyeon Kim, Nathanael Aubert-Kato, Nicolas Bredeche, Anthony J. Genot","doi":"10.1126/scirobotics.adr9578","DOIUrl":"https://doi.org/10.1126/scirobotics.adr9578","url":null,"abstract":"Swarm robotics relies on robots with minimal sensing and computation that achieve self-organization at the level of the collective to address a variety of tasks. Yet, swarm robotics still struggles with global tasks that require understanding the broader spatial context in which robots operate, such as identifying the shape of the arena in which they are embedded. Here, we introduce spectral swarm robotics, a framework in which robots recover global spectral descriptors of the swarm’s geometric structure through strictly local interactions. To do so, robots emulate diffusion to locally estimate Laplacian spectral invariants, quantities normally obtained only through centralized computation. These spectral values serve as compact descriptors of the environment. In the present proof of concept, robots estimated the second Laplacian eigenvalue, λ 2 , and used it to classify arenas from a predefined set of shapes with distinct spectral signatures. We show that estimating a single eigenvalue is sufficient for one-shot classification among selected confined environments with distinct spectral signatures. We uncovered a scaling law linking swarm size (a global parameter) and interaction radius (a local parameter) that ensures robust estimation. We validated the approach experimentally on two distinct hardware platforms, Kilobots and Pogobots, achieving reliable classification of arena shapes under realistic noise and sparse connectivity conditions. Last, we discuss how locally recovering Laplacian spectral invariants opens avenues for swarm robotics in settings where producing efficient collective behaviors strongly depends on the global structure of the swarm.","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"8 1","pages":""},"PeriodicalIF":25.0,"publicationDate":"2026-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149226143","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}

引用次数: 0

Fast Feature Field (F3): A general predictive representation of events for robot perception 快速特征域（f3）：机器人感知事件的一般预测表示

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-10-14 DOI: 10.1126/scirobotics.aec8218

Richeek Das, Kostas Daniilidis, Pratik Chaudhari

{"title":"Fast Feature Field (F3): A general predictive representation of events for robot perception","authors":"Richeek Das, Kostas Daniilidis, Pratik Chaudhari","doi":"10.1126/scirobotics.aec8218","DOIUrl":"10.1126/scirobotics.aec8218","url":null,"abstract":"<div >This paper develops a mathematical argument and algorithms for building representations of data from event-based cameras that we call Fast Feature Field (F<sup>3</sup>). We learned this representation by predicting future events from past events and show that it preserves scene structure and motion information. F<sup>3</sup> exploits the sparsity of event data and is robust to noise and variations in event rates. It can be computed efficiently using ideas from multiresolution hash encoding and deep sets, achieving 120 hertz at high-definition (HD) resolution and 440 hertz at video graphics array (VGA) resolution. F<sup>3</sup> represents events in a contiguous spatiotemporal volume as a multichannel image, enabling a range of downstream tasks. We obtained state-of-the-art performance on optical flow estimation, semantic segmentation, and monocular metric depth estimation, on data from three robotic platforms (a car, a quadruped robot, and a flying platform) and across different lighting conditions (daytime and nighttime), environments (indoors, outdoors, urban, and off-road), and dynamic vision sensors (resolutions and event rates). Our implementations can predict these tasks at 25 to 75 hertz at HD resolution.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 119","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149172574","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}

引用次数: 0

System design of the Ultra Mobility Vehicle: A driving, balancing, and jumping bicycle robot 超机动车辆的系统设计：一种驱动、平衡、跳跃的自行车机器人

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-10-14 DOI: 10.1126/scirobotics.aed3627

Benjamin Bokser, Daniel J. Gonzalez, Aaron Preston, Alex Bahner, Annika Wollschläger, Arianna Ilvonen, Asa Eckert-Erdheim, Ashwin Khadke, Bilal Hammoud, Dean Molinaro, Fabian Jenelten, Henry Mayne, Howie Choset, Igor Bogoslavskyi, Itic Tinman, James Tigue, Jan Preisig, Kaiyu Zheng, Kenny Sharma, Kim Ang, Laura Lee, Liana Margolese, Nicole Lin, Oscar Frias, Paul Drews, Ravi Boggavarapu, Rick Burnham, Samuel Zapolsky, Sangbae Kim, Scott Biddlestone, Sean Mayorga, Shamel Fahmi, Surya P. N. Singh, Tyler McCollum, Velin Dimitrov, William Moyne, Yu-Ming Chen, David Perry, Farbod Farshidian, Marco Hutter, Alfred Rizzi, Gabe Nelson

{"title":"System design of the Ultra Mobility Vehicle: A driving, balancing, and jumping bicycle robot","authors":"Benjamin Bokser, Daniel J. Gonzalez, Aaron Preston, Alex Bahner, Annika Wollschläger, Arianna Ilvonen, Asa Eckert-Erdheim, Ashwin Khadke, Bilal Hammoud, Dean Molinaro, Fabian Jenelten, Henry Mayne, Howie Choset, Igor Bogoslavskyi, Itic Tinman, James Tigue, Jan Preisig, Kaiyu Zheng, Kenny Sharma, Kim Ang, Laura Lee, Liana Margolese, Nicole Lin, Oscar Frias, Paul Drews, Ravi Boggavarapu, Rick Burnham, Samuel Zapolsky, Sangbae Kim, Scott Biddlestone, Sean Mayorga, Shamel Fahmi, Surya P. N. Singh, Tyler McCollum, Velin Dimitrov, William Moyne, Yu-Ming Chen, David Perry, Farbod Farshidian, Marco Hutter, Alfred Rizzi, Gabe Nelson","doi":"10.1126/scirobotics.aed3627","DOIUrl":"10.1126/scirobotics.aed3627","url":null,"abstract":"<div >Trials cyclists and mountain bike riders can hop, jump, balance, and drive on one or both wheels. This versatility allows them to achieve speed and energy efficiency on smooth terrain and agility over rough terrain. Inspired by these athletes, we present the design and control of a robotic platform, Ultra Mobility Vehicle (UMV), that combines a bicycle and a reaction mass to move dynamically with minimal actuated degrees of freedom. We used a simulation-driven design optimization process to synthesize a spatial linkage topology with a focus on vertical jump height and momentum-based balancing on a single wheel contact. Using a deep reinforcement learning (RL) framework, we demonstrate zero-shot transfer of diverse athletic behaviors, including track stands, jumps, wheelies, rear-wheel hopping, and front flips. This 23.5-kilogram robot is capable of high speeds (<span><math><msub><mi>v</mi><mtext>max</mtext></msub></math></span> = 8 meters per second) and jumping on and over large obstacles (1-meter tall or 130% of the robot’s nominal height).</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 119","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.aed3627","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149172572","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}

引用次数: 0

MonoRace: Robust monocular AI for champion-level drone racing MonoRace：用于冠军级无人机比赛的强大单目AI

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-10-14 DOI: 10.1126/scirobotics.aee9250

Stavrow A. Bahnam, Robin Ferede, Till M. Blaha, Anton E. Lang, Erin Lucassen, Quentin Missinne, Aderik E. C. Verraest, Christophe De Wagter, Guido C. H. E. de Croon

{"title":"MonoRace: Robust monocular AI for champion-level drone racing","authors":"Stavrow A. Bahnam, Robin Ferede, Till M. Blaha, Anton E. Lang, Erin Lucassen, Quentin Missinne, Aderik E. C. Verraest, Christophe De Wagter, Guido C. H. E. de Croon","doi":"10.1126/scirobotics.aee9250","DOIUrl":"10.1126/scirobotics.aee9250","url":null,"abstract":"<div >Autonomous drone racing represents a major frontier in robotics research. It requires an artificial intelligence (AI) that can run onboard lightweight flying robots under tight resource and time constraints while pushing the physical system to its limits. The state of the art in this area consists of a system with a stereo camera and an inertial measurement unit (IMU) that beat human drone-racing champions in a controlled indoor environment. Here, we present MonoRace, an onboard drone-racing approach that generalizes to a competition environment without any external motion tracking system, using only a monocular rolling-shutter camera and an IMU. The approach features robust state estimation that combines neural network–based gate segmentation with a drone model. Moreover, it includes an offline optimization procedure that leverages the known geometry of gates to refine any state estimation parameter. This offline optimization is based purely on onboard flight data and is important for fine-tuning the extrinsic camera calibration parameters. Furthermore, the guidance and control are performed by a neural network that forgoes inner-loop controllers by directly sending motor commands. This small network runs on the flight controller at 500 hertz. The proposed approach won the Abu Dhabi Autonomous Drone Racing Competition 2025, outperforming all competing autonomous drone race teams and three world champion human pilots in a direct knockout tournament. It reached speeds up to 100 kilometers per hour on the competition track and successfully coped with problems such as camera interference and IMU saturation.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 119","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.aee9250","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149172573","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}

引用次数: 0

ALBATROSS: A bioinspired, aerially deployable, autonomous sailing sensor platform 信天翁：一种受生物启发的、可在空中展开的自主航行传感器平台

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-09-30 DOI: 10.1126/scirobotics.aed1423

Shane Kyi Hla Win, Danial Sufiyan, Brian Leonard Suhadi, Luke Soe Thura Win, Xinyu Cai, Wei Jun Ang, Shaohui Foong

{"title":"ALBATROSS: A bioinspired, aerially deployable, autonomous sailing sensor platform","authors":"Shane Kyi Hla Win, Danial Sufiyan, Brian Leonard Suhadi, Luke Soe Thura Win, Xinyu Cai, Wei Jun Ang, Shaohui Foong","doi":"10.1126/scirobotics.aed1423","DOIUrl":"10.1126/scirobotics.aed1423","url":null,"abstract":"<div >Rapid deployment of autonomous sensor platforms in remote aquatic areas remains a challenge for environmental monitoring, disaster response, and ocean science. We present ALBATROSS (Airborne Lander with Buoyant AuToROtating Sailing Sensor), a hybrid bioinspired robot that passively autorotates for soft water landing and transitions into a wind-driven sailboat. Its dual-function rigid wingsails, acting as both sails and autorotating wings, enable efficient descent and propulsion. A freely rotating main sail, controlled by a tail rudder, and a directly actuated front sail enhance thrust and aerodynamic stability. A biomimetic caudal fin–inspired rudder enables propulsion during low-wind conditions and agile maneuvering in challenging upwind tacking. We empirically evaluated different wingsail airfoils through wind tunnel testing and optimized rudder design through hydrodynamic modeling, material selection, and speed-power efficiency analysis across different sizes and stiffnesses. Field experiments validate aerial deployment from 150 meters, autorotating descent, passive self-righting, and successful waypoint navigation using both rule-based control and a reinforcement learning approach. By integrating minimal actuation, bioinspired structures, and passive dynamics across aerial and marine operating regimes, this work extracts generalizable design principles for lightweight, aerially deployable sailing robots capable of robust, energy-efficient marine sensing.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 118","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.aed1423","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149077697","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}

引用次数: 0

Taking aerial-aquatic robots beyond biological capabilities 超越生物能力的空气-水机器人

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-09-30 DOI: 10.1126/scirobotics.ael4177

Raphael Zufferey, Andre Farinha

引用次数: 0

What if an alien believes our robot videos? 如果外星人相信了我们的机器人视频怎么办？

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-09-30 DOI: 10.1126/scirobotics.ael7878

Robin Murphy

引用次数: 0

Takeoff of an all-polymer micro aerial flying robot 全聚合物微型空中飞行机器人的起飞

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-09-30 DOI: 10.1126/scirobotics.aed8148

Yuan Zhu, Hanxiang Wu, Dawei Sun, William Budiman, Kede Liu, Yufeng Chen, Wenzhong Yan, Qibing Pei

{"title":"Takeoff of an all-polymer micro aerial flying robot","authors":"Yuan Zhu, Hanxiang Wu, Dawei Sun, William Budiman, Kede Liu, Yufeng Chen, Wenzhong Yan, Qibing Pei","doi":"10.1126/scirobotics.aed8148","DOIUrl":"10.1126/scirobotics.aed8148","url":null,"abstract":"<div >Flying insects are agile and can withstand impacts and compression because of their low inertia and resilient wings, exoskeletons, and muscles. These capabilities inspire the development of micro aerial vehicles (MAVs) for surveillance, disaster response, and environmental monitoring in confined or hazardous spaces. However, MAVs, especially subgram flapping-wing platforms, remain fragile because they rely on rigid components in their wings, transmissions, or actuators. We report a resilient all-polymer flying robot weighing 185 milligrams, powered by an electrostrictive bending actuator that directly drives cone-shaped compliant wings without a transmission. The actuator achieves a power density of 1600 watts per kilogram and a bending angle of 136° at 110 hertz, enabling the flying robot to achieve a lift-to-weight ratio of 3.0 and a lift-to-power ratio of 8.3 millinewtons per watt at 30 hertz. The robot demonstrates unaided takeoff with an average ascending speed of 40 centimeters per second. Thanks to its compliant and transmission-free structure, the robot can resume flight after being hit by a flyswatter or flattened by heavy loads. The robot design features mechanical simplicity and resilience, promising a paradigm for subgram flight in harsh environments.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 118","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149077695","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}

引用次数: 0

AI-driven actuation-compatible tracking for closed-loop navigation of miniature robots in vivo 微型机器人体内闭环导航的人工智能驱动驱动兼容跟踪

IF 25.5 1区计算机科学

Science Robotics Pub Date : 2026-09-23 DOI: 10.1126/scirobotics.aee0592

Delin Hu, Zhaoyang Qi, Yihang Jiang, Xurui Liu, Moqiu Zhang, Haojin Yang, Li Zhang

{"title":"AI-driven actuation-compatible tracking for closed-loop navigation of miniature robots in vivo","authors":"Delin Hu, Zhaoyang Qi, Yihang Jiang, Xurui Liu, Moqiu Zhang, Haojin Yang, Li Zhang","doi":"10.1126/scirobotics.aee0592","DOIUrl":"10.1126/scirobotics.aee0592","url":null,"abstract":"<div >Miniature wireless magnetic robots (MWMRs) hold immense potential for minimally invasive interventions. Real-time tracking of MWMRs is essential for precise in vivo navigation but remains challenging. Tracking MWMRs via their emitted magnetic signals offers a viable solution, yet it has been limited to centimeter-scale robots because isolating weak MWMR signals from overwhelming actuation-induced interference becomes increasingly difficult with miniaturization. Here, we present an artificial intelligence (AI)–driven tracking system to overcome these challenges, enabling precise localization of millimeter-scale MWMRs in the presence of actuation fields. Our approach integrates a sensor array design with a transformer to accurately model and filter the interference, followed by a spatiotemporal transformer that interprets the extracted MWMR signals over a short time window to corresponding three-dimensional coordinates. Our system has been validated in a range of in vitro and in vivo experiments, demonstrating its applicability to closed-loop navigation of MWMRs in unshielded clinical settings.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"11 118","pages":""},"PeriodicalIF":25.5,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149019899","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}

引用次数: 0