LIBERO-Safety: A Comprehensive Benchmark for Physical and Semantic Safety in Vision-Language-Action Models
For researchers developing safe VLA models, this benchmark provides a scalable pipeline and dataset to identify safety failures, though the findings are incremental as they confirm known tensions.
The paper introduces LIBERO-Safety, a benchmark for evaluating physical and semantic safety in Vision-Language-Action models, and provides a dataset of 19,664 collision-free demonstrations. Evaluation of eight VLA models reveals a tension between generalization and safety, with task success limited by trajectory synthesis and semantic misalignment.
Despite the impressive manipulation capabilities of Vision-Language-Action (VLA) models, their operational safety under strict constraints remains largely unverified. To address this, we introduce a parametric safety benchmark to procedurally generate safety-critical scenarios with comprehensive stochasticity. To overcome the scalability bottlenecks of human teleoperation, we develop a novel keypose-driven data generation pipeline. Leveraging this infrastructure, we curate a large-scale dataset of 19,664 strictly collision-free demonstrations with extensive domain randomization. We then conduct a systematic cross-paradigm evaluation of eight VLA and two embodied foundation models. Our analysis reveals a critical generalization-safety tension: although high-diversity training fosters safer trajectories, task success remains fundamentally bottlenecked by sub-optimal trajectory synthesis and semantic misalignment. By providing a scalable pipeline, a robust dataset, and profound failure-mode insights, LIBERO-Safety establishes a crucial foundation for developing safe and reliable VLA models.