Reproduce “Finetuning VLAs for Dexterous Manipulation: The Right Checkpoint Is All You Need” 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
Artur Kuramshin, Charlotte Cloutier, Michael Przystupa, Kirsty Ellis, Mohammad S. Akhter, Glen Berseth
The authors test whether fine-tuned robot models can be steered by reinforcement learning. They compare five recipes for π₀.₅ with a 16-degree-of-freedom hand in two simulated tasks. A separate learner chooses action-generating noise while the model stays fixed. On bulb twisting at 8k steps, EXPERT-ONLY reaches 100% on training placements and 88% on extrapolation placements; other recipes reach at most 33% extrapolation success. This is limited to tested placements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.