Reproduce “Correcting Persistent Failures in Point-Cloud Grasping Policies using Human Demonstrations in Virtual Reality” 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
Achyuthan Unni Krishnan, Karthik Rengarajan, Bharath Narasimhan, Chris von Csefalvay
The authors introduce targeted training for grasping unfamiliar objects. The robot flags repeated failures; a person proposes two contact points in VR, and the gripper verifies them before training, without copying hand motion. In simulation, success improves on demonstrated and hidden hard cases, but hidden performance remains near the warm start; bundled changes prevent attributing gains to human input alone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.