Simulation-based reinforcement learning for real-world autonomous driving
This addresses the challenge of reducing costs and engineering effort for real-world autonomous driving, though it is incremental as it builds on existing sim-to-real methods.
The authors tackled the problem of training an autonomous driving system using reinforcement learning in simulation and synthetic data, achieving successful sim-to-real policy transfer with a full-size real-world vehicle.
We use reinforcement learning in simulation to obtain a driving system controlling a full-size real-world vehicle. The driving policy takes RGB images from a single camera and their semantic segmentation as input. We use mostly synthetic data, with labelled real-world data appearing only in the training of the segmentation network. Using reinforcement learning in simulation and synthetic data is motivated by lowering costs and engineering effort. In real-world experiments we confirm that we achieved successful sim-to-real policy transfer. Based on the extensive evaluation, we analyze how design decisions about perception, control, and training impact the real-world performance.