Reproduce “ME-VLM: A Unified VLM for Embodied Cognition and Agent Coordination” 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
Foundation Model, Li Auto Inc
ME-VLM combines physical scene understanding with long-horizon planning, skill use, and feedback-guided correction in one vision-language model. It is trained through physical grounding, separate agent and embodied experts, and on-policy distillation into a single policy. ME-VLM 35B-A3B reports a 70.9 average across 26 embodied benchmarks, versus 62.7 for PhysBrain 1.5-8B; this aggregate does not isolate training contributions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.