Task-Driven Detection of Distribution Shifts with Statistical Guarantees for Robot Learning

TL;DR

This paper introduces a PAC-Bayes based method for task-driven out-of-distribution detection in robot learning, providing statistical guarantees on detection confidence and error rates, demonstrated in simulation and hardware tasks.

Contribution

It presents a novel PAC-Bayes framework for OOD detection with performance guarantees, tailored for robot tasks and sensitive to performance-impacting distribution changes.

Findings

Effective OOD detection with statistical guarantees in simulation and hardware.

Task-driven detection sensitive to environment changes affecting robot performance.

Rapid detection within few trials in real-world robotic tasks.

Abstract

Our goal is to perform out-of-distribution (OOD) detection, i.e., to detect when a robot is operating in environments drawn from a different distribution than the ones used to train the robot. We leverage Probably Approximately Correct (PAC)-Bayes theory to train a policy with a guaranteed bound on performance on the training distribution. Our idea for OOD detection relies on the following intuition: violation of the performance bound on test environments provides evidence that the robot is operating OOD. We formalize this via statistical techniques based on p-values and concentration inequalities. The approach provides guaranteed confidence bounds on OOD detection including bounds on both the false positive and false negative rates of the detector and is task-driven and only sensitive to changes that impact the robot's performance. We demonstrate our approach in simulation and hardware for a grasping task using objects with unfamiliar shapes or poses and a drone performing vision-based obstacle avoidance in environments with wind disturbances and varied obstacle densities. Our examples demonstrate that we can perform task-driven OOD detection within just a handful of trials.