SimScale: Learning to Drive via Real-World Simulation at Scale
This addresses the challenge of learning rational decisions in safety-critical and out-of-distribution scenarios for autonomous driving systems, representing a novel method for a known bottleneck.
The paper tackles the problem of insufficient data diversity for autonomous driving by introducing SimScale, a scalable simulation framework that synthesizes massive unseen states from existing driving logs, leading to significant improvements in robustness and generalization, such as +6.8 EPDMS on navhard and +2.9 on navtest benchmarks.
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 +6.8 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. Our simulation data and code would be released.