Reproduce “VLX-Go: Robust Embodied Tracking through Multi-Stage Learning and On-Policy Refinement” 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
Kyusong Lee, Heting Ying, Tiancheng Zhao
VLX-Go converts visual history and a language instruction into local robot-centric waypoints for a downstream controller. Its three-stage training aligns visual-language representations before specializing the policy for person tracking. On EVT-Bench, distractor-tracking success rises from 41.54% for OmTrackVLA to 55.80%, then to 58.22% after DAgger refinement. This supports the complete recipe, not isolated effects or a general refinement benefit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.