SimScale: Learning to Drive via Real-World Simulation at Scale

TL;DR

SimScale introduces a scalable simulation framework for autonomous driving that synthesizes diverse scenarios to improve policy robustness and generalization, leveraging neural rendering and pseudo-expert trajectories.

Contribution

The paper presents a novel simulation pipeline for generating unseen driving states and demonstrates its effectiveness in enhancing autonomous driving policies.

Findings

Significant performance improvements on real-world benchmarks (+8.6 EPDMS, +2.9 scores).

Policy improvements scale with increased simulated data without additional real data.

Design of pseudo-expert trajectories and analysis of scaling properties are crucial for success.

Abstract

Achieving fully autonomous driving systems requires learning rational decisions in a wide span of scenarios, including safety-critical and out-of-distribution ones. However, such cases are underrepresented in real-world corpus collected by human experts. To complement for the lack of data diversity, we introduce a novel and scalable simulation framework capable of synthesizing massive unseen states upon existing driving logs. Our pipeline utilizes advanced neural rendering with a reactive environment to generate high-fidelity multi-view observations controlled by the perturbed ego trajectory. Furthermore, we develop a pseudo-expert trajectory generation mechanism for these newly simulated states to provide action supervision. Upon the synthesized data, we find that a simple co-training strategy on both real-world and simulated samples can lead to significant improvements in both robustness and generalization for various planning methods on challenging real-world benchmarks, up to +8.6 EPDMS on navhard and +2.9 on navtest. More importantly, such policy improvement scales smoothly by increasing simulation data only, even without extra real-world data streaming in. We further reveal several crucial findings of such a sim-real learning system, which we term SimScale, including the design of pseudo-experts and the scaling properties for different policy architectures. Simulation data and code have been released at https://github.com/OpenDriveLab/SimScale.