End-to-End Imitation Learning with Safety Guarantees using Control Barrier Functions

TL;DR

This paper introduces a method for imitation learning that guarantees safety in complex systems by integrating Control Barrier Functions, ensuring robustness against disturbances in vision-based control tasks.

Contribution

It develops conditions for using safe expert controllers with CBFs to create end-to-end controllers that provide formal safety guarantees during learning.

Findings

Safety guarantees demonstrated in simulated inverted pendulum control.

Safety guarantees demonstrated in simulated car driving on a track.

IL with expert demonstrations achieves input-to-state safety under disturbances.

Abstract

Imitation learning (IL) is a learning paradigm which can be used to synthesize controllers for complex systems that mimic behavior demonstrated by an expert (user or control algorithm). Despite their popularity, IL methods generally lack guarantees of safety, which limits their utility for complex safety-critical systems. In this work we consider safety, formulated as set-invariance, and the associated formal guarantees endowed by Control Barrier Functions (CBFs). We develop conditions under which robustly-safe expert controllers, utilizing CBFs, can be used to learn end-to-end controllers (which we refer to as CBF-Compliant controllers) that have safety guarantees. These guarantees are presented from the perspective of input-to-state safety (ISSf) which considers safety in the context of disturbances, wherein it is shown that IL using robustly safe expert demonstrations results in ISSf with the disturbance directly related to properties of the learning problem. We demonstrate these safety guarantees in simulated vision-based end-to-end control of an inverted pendulum and a car driving on a track.