MuJoCo Manipulus: A Robot Learning Benchmark for Generalizable Tool Manipulation
Keywords: robotics, deep reinforcement learning, benchmark
TL;DR: We propose a Robot Learning Benchmark for Generalizable Tool Manipulation
Abstract: We propose MuJoCo Manipulus, a novel open-source benchmark powered by the MuJoCo physics simulation engine, designed to accelerate advances in robot learning for tool manipulation. Our benchmark includes a diverse set of tasks for tool manipulation --- a domain where the field currently lacks a unified benchmark. Different research groups rely on custom-designed tasks or closed-source setups, limiting cross-comparability and hindering significant progress in this field. To that end, our benchmark provides 16 challenging tool manipulation tasks, including variants of Pouring, Scooping, Scraping, Stacking, Gathering, Hammering, Mini-Golf, and Ping-Pong. The benchmark supports both state-based and vision-based observation spaces, is fully integrated with the Gymnasium API, and seamlessly connects with widely used Deep Reinforcement Learning libraries, ensuring easy adoption by the community. We conduct extensive reinforcement learning experiments on our benchmark, and our results demonstrate that there is substantial progress to be made for training tool manipulation policies. Our codebase and additional videos of the learned policies can be found on our anonymous project website: https://mujoco-manipulus.github.io
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13092
Loading