OpenDriveLab/SimScale

What it solves

SimScale addresses the challenge of training robust end-to-end autonomous driving planners by providing a scalable way to generate high-fidelity, reactive driving scenarios. It solves the data scarcity problem for critical driving events by synthesizing diverse simulation data that can be used to improve the generalization and robustness of driving models.

How it works

The project implements a scalable simulation pipeline that creates synthetic driving scenarios featuring "pseudo-expert" demonstrations. It employs a sim-real co-training strategy, where models are trained on both real-world data and synthesized simulation data. This co-training can be performed using either pseudo-expert demonstrations or by relying solely on rewards to guide the learning process, effectively bridging the gap between simulated and real-world environments.

Who it’s for

This tool is designed for researchers and engineers working on end-to-end autonomous driving, specifically those developing and training planners that need to be tested and validated against a wide variety of reactive driving scenarios.

Highlights

• Scalable Simulation Pipeline: Synthesizes high-fidelity reactive driving scenarios with pseudo-expert demonstrations.
• Sim-Real Co-training: A strategy that synergistically improves robustness and generalization across various end-to-end planners.
• Comprehensive Scaling Insights: Provides a recipe and analysis of the scaling properties of sim-real learning systems for autonomy.
• Diverse Model Support: Compatible with various backbones and agents like LTF, DiffusionDrive, and GTRS-Dense.