VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning
arXiv cs.AI · 7 Oct 2026, 03:50 AEST
VeriFine introduces a framework for scaling verification in self-improving policies, particularly in embodied reasoning. This approach addresses the limitations of fixed judges by co-evolving the policy, training curriculum, and judge for enhanced performance.
Key facts
- Self-improving policies continually expose new failure patterns, changing what their judges must be able to verify.
- We introduce VeriFine, an agent harness framework that scales verification through the co-evolution of the policy, training curriculum, and judge.
- Experiments on driving and robot navigation tasks demonstrate continuous self-improvement in both policy and judge capability across reinforcement and supervised fine-tuning.
Read aloud
VeriFine introduces a framework for scaling verification in self-improving policies, particularly in embodied reasoning. This approach addresses the limitations of fixed judges by co-evolving the policy, training curriculum, and judge for enhanced performance. Self-improving policies continually expose new failure patterns, changing what their judges must be able to verify. We introduce VeriFine, an agent harness framework that scales verification through the co-evolution of the policy, training curriculum, and judge. Experiments on driving and robot navigation tasks demonstrate continuous self-improvement in both policy and judge capability across reinforcement and supervised fine-tuning.
Sources in this cluster
- VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning
arXiv cs.AI · 7 Oct 2026, 03:50 AEST