VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning
VeriFine co-evolves an embodied policy, curriculum, and judge so verification keeps pace with new failure patterns.
VeriFine is an agent framework that scales verification by co-evolving an embodied policy, its training curriculum, and its judge. A Policy Improvement Loop uses a rubric judge to diagnose recurring failures and optimize the policy; when verification stalls, a Judge Improvement Loop seeks human guidance and coactively calibrates the judge. Experiments on driving and robot navigation show continued improvement of both policy and judge under reinforcement learning and supervised fine-tuning. The work targets spatial grounding, causal reasoning, and safety-aware decisions as failure patterns change.
- VeriFine co-evolves the policy, training curriculum, and verification judge.
- A rubric judge diagnoses failures and builds an adaptive curriculum.
- Humans calibrate the judge when verification becomes the bottleneck.
- Experiments cover driving and robot navigation under RL and supervised fine-tuning.
From the original article
1 min read at arxiv.orgSelf-improving policies continually expose new failure patterns, changing what their judges must be able to verify. However, current fixed judges constrain both optimization feedback and the discovery of useful training examples, limiting further self-improvement.
By Zewei Zhou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.