Distributionally Robust and Safe Imitation Learning

2026-07-15

AI Summary

Imitation learning is prone to performance degradation and safety risks under distributional shifts. This work proposes the first unified robust and safe imitation learning framework that jointly addresses policy-induced and uncertainty-induced distributional shifts by integrating Taylor Series Imitation Learning (TaSIL), distributionally robust adaptive control, and explicit safety constraints. The approach systematically embeds safety guarantees into a distributionally robust optimization formulation, thereby enhancing policy performance in uncertain environments while ensuring task safety. Experimental results demonstrate that the proposed framework effectively handles distributional shifts in drone navigation tasks, successfully completing missions while avoiding unsafe regions.

Abstract

Imitation learning (IL) has achieved remarkable success in complex decision-making tasks. However, its performance is highly sensitive to distribution shifts, which can pose significant safety risks. We propose a distributionally robust and safe IL framework that explicitly addresses both policy-induced and uncertainty-induced distribution shifts. Our approach develops a unified framework leveraging Taylor Series Imitation Learning (TaSIL) to mitigate policy-induced shifts and distributionally robust adaptive control to handle uncertainty-induced shifts. This architecture enables the formulation of an IL problem that optimizes performance under distributional uncertainty while systematically accounting for safety constraints. We demonstrate the effectiveness of the proposed approach on an unmanned aerial vehicle (UAV) case study where the UAV performs a task in an uncertain environment while avoiding unsafe regions.

Problem

• Imitation Learning
• Distribution Shift
• Safety
• Distributional Robustness
• Uncertainty

Innovation

• Distributionally Robust Imitation Learning
• Safe Imitation Learning
• Taylor Series Imitation Learning
• Adaptive Control
• Distribution Shift