VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning
VeriFine co-evolves policies, curricula, and judges so embodied agents keep self-improving as failure patterns change.
VeriFine is an agent harness that scales verification for self-improving embodied policies by co-evolving the policy, curriculum, and judge. A policy loop diagnoses recurring failures and builds an adaptive curriculum, while a judge loop uses selective human feedback and coactive calibration when verification stalls. Experiments on driving and robot navigation show continued gains in both policy and judge quality under reinforcement learning and supervised fine-tuning.
- Policy loop uses a rubric judge to build an adaptive curriculum.
- Judge loop seeks human guidance when verification becomes the bottleneck.
- Coactive calibration aligns humans and agents on physical-reasoning rubrics.
- Driving and navigation experiments cover reinforcement and supervised fine-tuning.
From the original article
1 min read at huggingface.co
Self-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, Rachel Luo, Yulong Cao · huggingface.co
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.