Reproduce “GoalDex: Learning Image-Goal-Conditioned Policies for Dexterous In-Hand Manipulation” 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
Arjun Bhardwaj, Mayank Mittal, Maximum Wilder-Smith, Vaishakh Patil, Marco Hutter
GoalDex enables dexterous reorientation from images and joint feedback with a policy trained in simulation. At deployment, it uses current RGB views and goal images without an online object-pose estimate. Across five tested objects, it completed 80/100 trials versus 27/100 for a pose-based baseline. Success required a completion prediction and true orientation error below 0.4 radians. Geometry and masks remain necessary; unseen-category generalization is untested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.