Event
AFIL: Guiding VLA Policy Training with Failure Trajectories
1 reports · 1 sources
Hanwen Ren and other authors propose Adaptive Failure-Guided Learning (AFIL), a framework that treats failure trajectories as adaptive negative guidance for diffusion and flow-based VLA policies. The authors state that AFIL uses a pretrained VLA to generate failure trajectories online, eliminating the need to manually design failure modes or human intervention for recovery; it jointly trains two action generators (DAG), one for success and one for failure, on a shared vision-language backbone, and at sampling time adjusts the guidance strength according to the distance between the success and failure distributions at each diffusion step. The authors report that on in-domain and out-of-domain, short-horizon and long-horizon robotic manipulation tasks, AFIL improves task success rate and robustness compared with existing VLA baselines. These results come from the paper's experiments and await independent verification.
Report timeline
- Tokens of Motion: Primitive Motion Learning for Human-Humanoid Skill Transfer
The paper proposes the Adaptive Failure-Informed Learning (AFIL) framework, which uses a pretrained VLA to generate failure trajectories online as adaptive negative guidance for diffusion- and flow-based VLA policies, avoiding the need for handcrafted failure modes or human intervention for recovery. The method jointly trains two action generators (DAGs) for success and failure on a shared vision-language backbone, and at sampling time adjusts the guidance strength according to the distance between the success and failure distributions at each diffusion step. On in-domain and out-of-domain, short-horizon and long-horizon robotic manipulation tasks, AFIL improves task success rate and robustness compared with existing VLA baselines.
arXiv cs.RO ·