Robotics: Science and Systems 2025 Awards Announced! Leading Research Led by Pieter Abbeel Wins Outstanding Demo Award

Congratulations to the winners.

RSS (Robotics: Science and Systems) is a top-tier robotics conference held annually since 2005, dedicated to advancing scientific research and technological applications in robotics.

This year's conference took place from June 21 to 25 in Los Angeles, USA. Multiple awards, including the Outstanding Demo Award, Outstanding System Paper Award, Outstanding Student Paper Award, and Outstanding Paper Award, have been announced.

Link to awards page: https://roboticsconference.org/program/awards/

Outstanding Demo Paper Award

Title: Demonstrating MuJoCo Playground

• Paper link: https://www.roboticsproceedings.org/rss21/p020.pdf
• Project homepage: https://playground.mujoco.org/
• Institutions: UC Berkeley, Google DeepMind, University of Toronto, University of Cambridge
• Authors: Kevin Zakka, Baruch Tabanpour, Qiayuan Liao, Mustafa Haiderbhai, Samuel Holt, Jing Yuan Luo, Arthur Allshire, Erik Frey, Koushil Sreenath, Lueder Alexander Kahrs, Carmelo Sferrazza, Yuval Tassa, Pieter Abbeel

Abstract: This research introduces MuJoCo Playground, an open-source robot learning framework built on MJX, designed to simplify simulation setup, model training, and transfer from simulation to real-world applications. It supports quadruped robots, humanoids, dexterous hands, and robotic arms, enabling zero-shot sim-to-real transfer based on state or pixel inputs. The framework relies on integrated physics engines, batch renderers, and training environment tech stacks.

      ^{Quadruped robot}

      ^{Humanoid robot}

      ^{Robot falling and recovering}

Outstanding System Paper Award

Title: Building Rome with Convex Optimization

• Paper link: https://arxiv.org/pdf/2502.04640
• Project page: https://computationalrobotics.seas.harvard.edu/XM/
• Institution: Harvard University
• Authors: Haoyu Han, Heng Yang

Abstract: This study proposes SBA (scaled bundle adjustment), a method that uses learned depth to lift 2D keypoints to 3D, along with a convex SDP relaxation for global optimality. It employs CUDA-based trust-region optimizer (XM) to solve large-scale SDP relaxations, forming an efficient SfM pipeline that outperforms existing methods in quality, speed, and scalability.

      ^{XM optimizer solving 10,155 frames in one hour}

      ^{Reconstructed scene (left) vs. input image from Replica dataset (right)}

Outstanding Student Paper Award

Title: Solving Multi-Agent Safe Optimal Control with Distributed Epigraph Form MARL

• Paper link: https://arxiv.org/pdf/2504.15425
• Project page: https://mit-realm.github.io/def-marl/
• Institution: Harvard University
• Authors: Songyuan Zhang, Oswin So, Mitchell Black, Zachary Serlin, Chuchu Fan

Abstract: This work addresses multi-robot coordination with safety constraints, formalized as a constrained Markov decision process (CMDP). It introduces Def-MARL, a centralized training, decentralized execution framework using convex epigraph formulations to ensure zero constraint violations, validated through simulations and real robot experiments.

      ^{Multi-robot collaboration through narrow corridors}

      ^{Collaboration check target}

Test of Time Award

Additionally, the RSS 2025 Test of Time Award was announced: the 2009 paper by Nathan Michael, Jonathan Fink, and Vijay Kumar titled Cooperative Manipulation and Transportation with Aerial Robots.

Paper link: https://arxiv.org/pdf/2502.04640

Project page: https://roboticsconference.org/program/testoftimeaward/