Learning robust driving policies without online exploration

TL;DR

This paper introduces a multi-time-scale predictive representation learning approach for offline reinforcement learning, enabling robust driving policies that generalize to new road conditions without online exploration.

Contribution

The paper presents a novel offline learning method that improves generalization and robustness of driving policies in unseen environments, reducing reliance on online exploration.

Findings

Effective in generalizing to novel road geometries

Robust to damaged and distracting lane conditions

Performs well in both simulation and real-world tests

Abstract

We propose a multi-time-scale predictive representation learning method to efficiently learn robust driving policies in an offline manner that generalize well to novel road geometries, and damaged and distracting lane conditions which are not covered in the offline training data. We show that our proposed representation learning method can be applied easily in an offline (batch) reinforcement learning setting demonstrating the ability to generalize well and efficiently under novel conditions compared to standard batch RL methods. Our proposed method utilizes training data collected entirely offline in the real-world which removes the need of intensive online explorations that impede applying deep reinforcement learning on real-world robot training. Various experiments were conducted in both simulator and real-world scenarios for the purpose of evaluation and analysis of our proposed claims.