Summary
It’s hard to train self-driving cars in simulation and make them work well in the real world, especially on rough, unstructured roads. The researchers created MILER, a system that learns from a special intermediate map-like view combining camera and LiDAR data, so the car understands both simulated and real environments similarly. Instead of controlling the car directly, MILER plans safe paths matching what it learned in simulation, allowing it to drive real test tracks fully autonomously without needing extra real-world training. They demonstrated this by driving cars on a challenging 3 km track multiple times with no human help.
What this means in practice
- •For autonomous vehicle engineers: Use semantic mid-level representations to enable real-world driving from simulation-trained reinforcement learning policies without extra real-world tuning.
- •For robotic navigation teams: Integrate multi-sensor semantic bird's-eye views into control systems to improve deployment of simulation-trained navigation policies in unstructured environments.
Authors
Thomas Steinecker, Denis Trescher, Alexander Bienemann, Thorsten Luettel, Mirko Maehlisch
Abstract
Reinforcement learning constitutes a promising approach owing to its potential for superhuman performance and self-learned policies. However, its application to real-world autonomous driving remains scarce, particularly in unstructured environments, because of the challenges associated with sim-to-real transfer for unstructured environments. In this work, we present MILER, an end-to-end policy framework with zero-shot sim-to-real transfer. During offline training, we employ a custom semantic mid-level representation (MLR) simulator and train the policy network using reinforcement learning, with its control outputs applied directly to a bicycle model. During deployment on the real vehicle, camera and LiDAR data are processed by BEVFusion to generate a semantic bird's-eye-view representation consistent with that of the MLR simulator. The actions generated by the policy network are not applied directly to the real vehicle. Instead, we employ a trajectory-alignment strategy that enables zero-shot sim-to-real transfer of both perception and control. We extensively evaluate the proposed framework on a diverse test track comprising numerous challenges, including various obstacles, hairpin curves, velocities of up to 33.6 km/h, and off-road sections. In total, we drove 17.3 km with two different vehicles on a 3.0 km test track without human intervention, thereby demonstrating the effectiveness of our approach. Furthermore, the entire software stack runs on a Jetson AGX Orin.