ICML: Imitation, Prediction, and Model-Based Reinforcement Learning for Autonomous Driving

Thanks to @kargarisaac for sharing this on Twitter.

1 Like

Proof of concept of an imitation learning-based approach in the CARLA driving simulator:

This “Deep Imitative Model” approach learned from 25 hours of driving using CARLA’s autopilot (not human driving). The results are based off of 7500 test scenarios that last 100 seconds each.

Paper: