Guardian: Detecting Robotic Planning and Execution Errors with Vision-Language Models
Published in CoRL 2025 Workshop Robot Data, 2025
TLDR; Guardian introduces an automatic robot failure synthesis approach that generates diverse planning and execution failures with fine-grained reasoning traces. We train a VLM that achieves state-of-the-art performance on failure detection benchmarks and effectively improves task success rates in both simulation and real robots.
Recommended citation: Paul Pacaud, Ricardo Garcia, Shizhe Chen, and Cordelia Schmid https://arxiv.org/abs/2512.01946