Reproduce “WetRobo: A Reproducible Robot Kit for Coding Agents in Biological Laboratories” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yuna Oikawa, Kei Endo, Takanori Uzawa, Yunzhe Zhang, Manan Anjaria, Lerrel Pinto, Sherry Yang, Koji Tsuda
WetRobo presents a reproducible robot kit that lets coding agents adapt manipulation programs to the equipment, cameras, lighting, and geometry of individual biological laboratories without collecting new teleoperation data or retraining a neural policy. Across Petri-dish, bottle-cap, and incubator-door tasks, the approach achieved 10/10 repeated bottle-cap trials in two laboratories, while a π₀.₅ policy fine-tuned in one lab achieved 10/10 there but 0/10 after transfer.
Key insights
- Code-as-policy shifts adaptation from neural-network fine-tuning to inspectable program generation, letting the agent explicitly incorporate local calibration, collision checks, grasp tests, and completion verification.
- The same kit can produce different control strategies for different laboratories: for example, AprilTags and MuJoCo checks in one setting, versus RGB-D geometry and image Jacobians in another.
- Feedback from failed grasps and incomplete motions can be converted into reusable procedures, creating an incremental repository of laboratory-specific robot knowledge.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.