Activity stream from the Prorok Lab — papers, talks, videos, posts, and openings.https://www.proroklab.org/https://www.proroklab.org/posts/open-day/https://www.proroklab.org/posts/open-day/Teams of small robots interpreted instructions given in plain English and autonomously carried out complex missions — staying synchronised, avoiding collisions, and adapting in real time to unexpected disruptions.Tue, 24 Mar 2026 00:00:00 GMThttps://www.proroklab.org/posts/bbc-click-fantastico/https://www.proroklab.org/posts/bbc-click-fantastico/This work bridges CONtinuous robot dynamics and scalable disCRETE search — what we call "concrete" planning. Our approach enables robots to execute agile, safe, and efficient motions even in dense environments.Wed, 18 Mar 2026 00:00:00 GMThttps://www.proroklab.org/publications/concrete-mapf-npj-robotics/https://www.proroklab.org/publications/concrete-mapf-npj-robotics/PAPER — Keisuke Okumura, Guang Yang, Zhan Gao, Heedo Woo, Amanda ProrokMon, 16 Mar 2026 00:00:00 GMThttps://www.proroklab.org/videos/concrete-mapf/https://www.proroklab.org/videos/concrete-mapf/Framework for aggressive multi-robot coordination deployed with 40 robots (20 aerial, 8 ground, 12 obstacles) in a compact lab space. Published in npj Robotics (2026).Mon, 16 Mar 2026 00:00:00 GMThttps://www.proroklab.org/posts/cambridge-festival-visit/https://www.proroklab.org/posts/cambridge-festival-visit/Researchers in our lab will be showcasing a system that lets teams of robots carry out instructions given in plain English — listening to human voice, then completing their mission completely on their own.Thu, 12 Mar 2026 00:00:00 GMThttps://www.proroklab.org/posts/cambridge-academic-life/https://www.proroklab.org/posts/cambridge-academic-life/The Prorok Lab has grown so much in the last 8 years and continues to be a team driven by collaboration.Tue, 10 Mar 2026 00:00:00 GMThttps://www.proroklab.org/posts/iclr2026-accepted/https://www.proroklab.org/posts/iclr2026-accepted/Work that spans robot policy watermarking, higher-order interactions in multi-agent pathfinding, and the role of diversity in MARL — covering remote auditing of robot policies, hypergraph attention for dense MAPF, and when heterogeneity actually pays off.Tue, 10 Feb 2026 00:00:00 GMThttps://www.proroklab.org/posts/ieee-ras-feature/https://www.proroklab.org/posts/ieee-ras-feature/Matteo's thesis on neural diversity in multi-agent learning provides a comprehensive and impactful study, demonstrating that diversity is a previously under-explored yet fundamental factor for effective collective learning in MARL.Fri, 06 Feb 2026 00:00:00 GMThttps://www.proroklab.org/posts/aaai2026-pathfinding/https://www.proroklab.org/posts/aaai2026-pathfinding/Big congratulations to Rishabh Jain, Keisuke Okumura and Michael Amir on presenting "Graph Attention-Guided Search for Dense Multi-Agent Pathfinding" at AAAI 2026 — pushing forward the state of the art in scalable multi-agent coordination.Tue, 27 Jan 2026 00:00:00 GMThttps://www.proroklab.org/publications/iclr2026-diversity-rewarded/https://www.proroklab.org/publications/iclr2026-diversity-rewarded/PAPER — Michael Amir, Matteo Bettini, Amanda ProrokTue, 20 Jan 2026 00:00:00 GMThttps://www.proroklab.org/publications/iclr2026-hypergraph-mapf/https://www.proroklab.org/publications/iclr2026-hypergraph-mapf/PAPER — Rishabh Jain, Keisuke Okumura, Michael Amir, Pietro Liò, Amanda ProrokTue, 20 Jan 2026 00:00:00 GMThttps://www.proroklab.org/publications/iclr2026-watermarking/https://www.proroklab.org/publications/iclr2026-watermarking/PAPER — Michael Amir, Manon Flageat, Amanda ProrokTue, 20 Jan 2026 00:00:00 GMThttps://www.proroklab.org/posts/multiagent-systems-feature/https://www.proroklab.org/posts/multiagent-systems-feature/Grateful for a brilliant 2025 and excited for what lies ahead in 2026 — here's to continued growth, new opportunities, and building what's next.Mon, 12 Jan 2026 00:00:00 GMThttps://www.proroklab.org/publications/neurips2025-thompson-sampling/https://www.proroklab.org/publications/neurips2025-thompson-sampling/PAPER — Jasmine Bayrooti, Sattar Vakili, Amanda Prorok, Carl Henrik EkWed, 10 Dec 2025 00:00:00 GMThttps://www.proroklab.org/posts/jasmine-bayrooti-joins/https://www.proroklab.org/posts/jasmine-bayrooti-joins/Jasmine Bayrooti will be presenting work that establishes no-regret guarantees for Thompson sampling in episodic Markov Decision Processes modeled via joint multi-output Gaussian Processes.Tue, 02 Dec 2025 00:00:00 GMThttps://www.proroklab.org/posts/continual-learning-robotics/https://www.proroklab.org/posts/continual-learning-robotics/Bringing together diverse teams, perspectives, and research agendas sparked exactly the kind of conversations we need as robotics moves rapidly from controlled environments into the dynamic, unpredictable real world.Wed, 19 Nov 2025 00:00:00 GMThttps://www.proroklab.org/posts/aaai2026-mapf/https://www.proroklab.org/posts/aaai2026-mapf/For (arguably) the first time, our neural approach — LaGAT — clearly surpasses leading search-based planners. The key: fusing GNN-based imitation learning with the LaCAM search planner to get the best of both worlds.Mon, 17 Nov 2025 00:00:00 GMThttps://www.proroklab.org/publications/lagat-aaai26/https://www.proroklab.org/publications/lagat-aaai26/PAPER — Rishabh Jain, Keisuke Okumura, Michael Amir, Amanda ProrokSat, 08 Nov 2025 00:00:00 GMThttps://www.proroklab.org/posts/science-comms-impact/https://www.proroklab.org/posts/science-comms-impact/On air we discussed how our lab is pushing the boundaries of collective intelligence in multi-robot and multi-agent systems — and what it means to take this work to a broader audience.Mon, 27 Oct 2025 00:00:00 GMThttps://www.proroklab.org/posts/iros2025-presence/https://www.proroklab.org/posts/iros2025-presence/D4orm introduces a novel optimization framework for multi-robot trajectory generation. Inspired by diffusion generative models, it iteratively denoises noisy trajectories into smooth, collision-free, dynamically feasible ones — ready for real-world deployment.Tue, 14 Oct 2025 00:00:00 GMThttps://www.proroklab.org/posts/hiring-ra-vision/https://www.proroklab.org/posts/hiring-ra-vision/We're looking for a motivated Postdoc to develop and deploy computer vision systems on real robots — from tracking and 3D reconstruction to visual SLAM and in-field testing with aerial and ground platforms.Thu, 09 Oct 2025 00:00:00 GMThttps://www.proroklab.org/posts/corl2025-presence/https://www.proroklab.org/posts/corl2025-presence/Michael Amir presented ReCoDe at the poster session, Peter Woo showcased the Sanity quadrotor at the Open Source Hardware Workshop, and our alumnus Steven Morad delivered a spotlight talk at the RemembRL Workshop.Wed, 01 Oct 2025 00:00:00 GMThttps://www.proroklab.org/posts/smart-city-robotics/https://www.proroklab.org/posts/smart-city-robotics/Our winning solution leverages our expertise in multi-agent systems and perception, utilizing an off-board camera network to provide the robot with real-time contextual information — demonstrating the power of collective intelligence in a dynamic, social setting.Mon, 29 Sep 2025 00:00:00 GMThttps://www.proroklab.org/posts/extending-robot-minds-collective/https://www.proroklab.org/posts/extending-robot-minds-collective/The article champions a paradigm shift that lies at the heart of our lab's mission: moving away from building do-it-all robots, and instead advancing collective robotic intelligence — diverse, specialized robots that learn, adapt, and collaborate.Thu, 25 Sep 2025 00:00:00 GMThttps://www.proroklab.org/publications/extending-robot-minds/https://www.proroklab.org/publications/extending-robot-minds/PAPER — Amanda ProrokWed, 24 Sep 2025 00:00:00 GMThttps://www.proroklab.org/posts/science-robotics-extending-minds/https://www.proroklab.org/posts/science-robotics-extending-minds/I argue for a new paradigm: collective robotic intelligence. Instead of a single, central "brain", we should design robot collectives made up of diverse, specialized agents that learn and work together.Wed, 24 Sep 2025 00:00:00 GMThttps://www.proroklab.org/posts/corl2025-marl/https://www.proroklab.org/posts/corl2025-marl/ReCoDe introduces a reinforcement learning framework that dynamically adapts constraints, enabling agents to coordinate more effectively in challenging environments.Tue, 16 Sep 2025 00:00:00 GMThttps://www.proroklab.org/publications/recode-corl2025/https://www.proroklab.org/publications/recode-corl2025/PAPER — Michael Amir, Guang Yang, Zhan Gao, Keisuke Okumura, Heedo Woo, Amanda ProrokTue, 12 Aug 2025 00:00:00 GMThttps://www.proroklab.org/videos/recode-corl2025/https://www.proroklab.org/videos/recode-corl2025/Hybrid decentralized framework combining optimization-based control with MARL. Learns dynamic constraints to improve multi-agent coordination. Presented at CoRL 2025.Tue, 12 Aug 2025 00:00:00 GMThttps://www.proroklab.org/posts/multi-drone-systems/https://www.proroklab.org/posts/multi-drone-systems/At just 72g with brushless motors, onboard WiFi, and dual processors, Sanity hits speeds over 6 m/s and supports large-scale experiments with up to 20 drones flying simultaneously indoors.Thu, 10 Jul 2025 00:00:00 GMThttps://www.proroklab.org/publications/d4orm-iros2025/https://www.proroklab.org/publications/d4orm-iros2025/PAPER — Yixiao Zhang, Keisuke Okumura, Heedo Woo, Ajay Shankar, Amanda ProrokFri, 04 Jul 2025 00:00:00 GMThttps://www.proroklab.org/videos/d4orm-iros2025/https://www.proroklab.org/videos/d4orm-iros2025/Optimization method generating kinodynamically feasible, collision-free multi-robot trajectories using diffusion denoising. Evaluated for teams of up to 16 robots in 2D and 3D worlds. Presented at IROS 2025.Fri, 04 Jul 2025 00:00:00 GMThttps://www.proroklab.org/posts/phd-graduation-2025/https://www.proroklab.org/posts/phd-graduation-2025/Three incredible members of the Prorok Lab — Steven Morad, Jan Blumenkamp and Ryan Kortvelesy — have recently graduated with their PhDs. Their dedication and brilliance have been a huge part of our lab's journey.Tue, 17 Jun 2025 00:00:00 GMThttps://www.proroklab.org/posts/rldm2025-marl/https://www.proroklab.org/posts/rldm2025-marl/The talk highlighted why behavioral diversity is more than a design choice — it's a necessity for achieving complex, high-level coordination in large robot teams.Mon, 16 Jun 2025 00:00:00 GMThttps://www.proroklab.org/posts/icra2025-offline-rl/https://www.proroklab.org/posts/icra2025-offline-rl/Instead of using LLMs to directly control robots, we use them to translate human language into a compact, consistent latent vector that captures the essence of the command — and our offline-trained robots act on that vector regardless of how the instruction is phrased.Fri, 23 May 2025 00:00:00 GMThttps://www.proroklab.org/posts/multi-robot-slam/https://www.proroklab.org/posts/multi-robot-slam/DVM-SLAM enables multiple robots to collaboratively map unknown environments while simultaneously tracking their positions — using only low-cost, lightweight monocular cameras and operating in a fully decentralized manner.Wed, 21 May 2025 00:00:00 GMThttps://www.proroklab.org/publications/dvm-slam-icra2025/https://www.proroklab.org/publications/dvm-slam-icra2025/PAPER — Joshua Bird, Jan Blumenkamp, Amanda ProrokThu, 15 May 2025 00:00:00 GMThttps://www.proroklab.org/publications/language-conditioned-offline-rl-icra2025/https://www.proroklab.org/publications/language-conditioned-offline-rl-icra2025/PAPER — Steven Morad, Ajay Shankar, Jan Blumenkamp, Amanda ProrokThu, 15 May 2025 00:00:00 GMThttps://www.proroklab.org/publications/iclr2025-thompson-sampling/https://www.proroklab.org/publications/iclr2025-thompson-sampling/PAPER — Jasmine Bayrooti, Carl Henrik Ek, Amanda ProrokThu, 24 Apr 2025 00:00:00 GMThttps://www.proroklab.org/posts/efficient-model-based-rl/https://www.proroklab.org/posts/efficient-model-based-rl/HOT-GP is a model-based reinforcement learning algorithm that explores efficiently by reasoning about joint uncertainty over both environment dynamics and rewards — sampling transitions conditioned on optimistic rewards to imagine plausible, high-value futures.Thu, 24 Apr 2025 00:00:00 GMThttps://www.proroklab.org/posts/welcome-research-team/https://www.proroklab.org/posts/welcome-research-team/Manon brings a wealth of expertise in diversity-seeking machine learning and reinforcement learning algorithms — and we're thrilled to have her on board.Wed, 23 Apr 2025 00:00:00 GMThttps://www.proroklab.org/talks/ace-network-forum-oxford-2025/https://www.proroklab.org/talks/ace-network-forum-oxford-2025/Invited · ACE Network Forum — University of OxfordWed, 26 Mar 2025 00:00:00 GMThttps://www.proroklab.org/posts/ace-network-forum-oxford/https://www.proroklab.org/posts/ace-network-forum-oxford/Amanda shared our lab's latest research on how multi-agent systems can enable smarter decision-making — whether through cooperative robot perception or optimizing human-led fleet control.Wed, 26 Mar 2025 00:00:00 GMThttps://www.proroklab.org/posts/bbc-click-feature/https://www.proroklab.org/posts/bbc-click-feature/From swarm robotics to autonomous collaboration, our innovations are shaping the future of intelligent systems.Mon, 17 Mar 2025 00:00:00 GMThttps://www.proroklab.org/publications/co-optimizing-environments-tro2025/https://www.proroklab.org/publications/co-optimizing-environments-tro2025/PAPER — Zhan Gao, Guang Yang, Amanda ProrokSat, 01 Mar 2025 00:00:00 GMThttps://www.proroklab.org/posts/hiring-ra-natural-language/https://www.proroklab.org/posts/hiring-ra-natural-language/The Prorok Lab is looking for a talented Research Assistant in Adaptive Multi-Robot Mission Control via Natural Language — at the intersection of LLMs, NLP, and multi-robot systems.Tue, 04 Feb 2025 00:00:00 GMThttps://www.proroklab.org/openings/phd-perception-aerial-swarms/https://www.proroklab.org/openings/phd-perception-aerial-swarms/Develop robot policies that respond to high-level natural language commands at the intersection of LLMs, NLP, and multi-robot systems. Collaborate with students and researchers on cutting-edge projects in a world-class research environment.Tue, 04 Feb 2025 00:00:00 GMThttps://www.proroklab.org/publications/benchmarl-jmlr2024/https://www.proroklab.org/publications/benchmarl-jmlr2024/PAPER — Matteo Bettini, Amanda Prorok, Vincent MoensThu, 19 Dec 2024 00:00:00 GMThttps://www.proroklab.org/posts/neurips2024-benchmarl/https://www.proroklab.org/posts/neurips2024-benchmarl/BenchMARL is a library designed to simplify and standardize benchmarking for Multi-Agent Reinforcement Learning — letting researchers seamlessly mix and match MARL algorithms, tasks, and models with rigorous reproducibility.Thu, 19 Dec 2024 00:00:00 GMThttps://www.proroklab.org/talks/neurips2024-benchmarl/https://www.proroklab.org/talks/neurips2024-benchmarl/Conference · NeurIPS 2024Thu, 12 Dec 2024 00:00:00 GMT