To Do or Not to Do: Ensuring the Safety of Visuomotor Policies Learned from Demonstrations

2026-05-01

AI Summary

This work addresses the frequent neglect of runtime safety in existing vision-based imitation learning methods, which often prioritize task success at the expense of safe execution. The authors propose a policy-agnostic safety metric termed “execution assurance,” which identifies safe regions in the state space via view synthesis. Leveraging Nagumo’s sub-tangential condition and set invariance theory, they rigorously prove that any policy operating within these regions is guaranteed to execute safely while achieving optimal task success. By introducing execution assurance into visual imitation learning for the first time, this approach not only enables verifiably safe policy deployment but also yields a recovery strategy that mitigates the safety–performance trade-off. Simulations and real-world experiments on a Franka robot demonstrate that the method ensures diverse imitation policies attain maximal task success within the safe region, significantly enhancing overall performance.

Abstract

Task success has historically been the primary measure of policy performance in imitation learning (IL) research. This characteristics strictly limits the ubiquitous applications of IL algorithms in field robotics where safety assurance, in addition to task-success, is of paramount importance. It is often desirable for an IL-powered robot in the field not to roll out a policy, and hence score a poor performance, if the safety is not guaranteed. Although this trade-off between safety and performance is well investigated in classical control literature, policy safety is a heavily underexplored domain in IL research. There is no universal definition of safety in IL. To make things worst, many existing theoretical works on safety is notoriously difficult to extend to IL-powered robots in the field. This paper offers important insights on the safety and performance of IL policies. We propose execution guarantee, a policy-agnostic safety measure that guarantees the maximum task success for a visuomotor IL policy, despite minor run-time changes, from within a specific region in the state space. We leverage recent advances in view synthesis to identify such regions in the state space for an IL policy and explore a fundamental result on set invariance - namely, Nagumo's sub-tangentiality condition - to prove and operationalize execution guarantee from inside that region. Experiments with a Franka robot, both in simulation and real world, demonstrate how the proposed safety analysis allows various IL policies to achieve maximum task success with guarantee. We also demonstrate some interesting results on how a recovery policy - a by-product of the proposed safety analysis - can help to increase the policy performance and thereby mitigating the safety-performance tradeoff in IL.

Problem

• Imitation Learning
• Policy Safety
• Visuomotor Policies
• Safety-Performance Tradeoff
• Execution Guarantee

Innovation

• execution guarantee
• imitation learning safety
• view synthesis
• set invariance
• Nagumo's condition