Reward (Mis)design for Autonomous Driving

TL;DR

This paper introduces eight sanity checks to diagnose common errors in reward function design for autonomous driving, revealing widespread flaws and suggesting future research directions for improved reward design.

Contribution

It develops simple diagnostic checks for reward functions and applies them to autonomous driving, uncovering pervasive flaws and guiding future reward design improvements.

Findings

Widespread flaws found in reward functions for autonomous driving

Sanity checks effectively identify reward design errors

Guidelines proposed for future reward function development

Abstract

This article considers the problem of diagnosing certain common errors in reward design. Its insights are also applicable to the design of cost functions and performance metrics more generally. To diagnose common errors, we develop 8 simple sanity checks for identifying flaws in reward functions. These sanity checks are applied to reward functions from past work on reinforcement learning (RL) for autonomous driving (AD), revealing near-universal flaws in reward design for AD that might also exist pervasively across reward design for other tasks. Lastly, we explore promising directions that may aid the design of reward functions for AD in subsequent research, following a process of inquiry that can be adapted to other domains.