This month was devoted to produce results and write RLAD: Reinforcement Learning from Pixels for Autonomous Driving in Urban Environments. This paper was submitted to ICCV 2023.
Additionally, a video was edited to accompany the document as supplementary material:
What's Next?
RLAD addresses the challenge of training the encoder and policy network simultaneously using model-free Reinforcement Learning methods. The next steps will be focused on using model-based RL methods. Adapting Dreamer-V3 to the domain of autonomous driving would be a good starting point.