Video demonstration of: Model-based versus Model-free Deep Reinforcement Learning for Autonomous Racing Cars In this experiment, we evaluate Dreamer policies with respect to their sim2real transferability. We test trained dreamers deployed on the car in a physical test-track. This is a video demonstrating the driving performance of the Dreamer agent. It presents the Dreamer agent completing a full lap in TRT in the forward direction. Then, to evaluate its generalization, we tested the agent in reversed direction. We observe that even if the agent was not trained in the reversed direction and not directly on the TRT track, the agent can complete a successful lap. Finally, we placed two obstacles after the most challenging turn and ran the agent in this configuration. The Dreamer agents can complete the lap and securely avoid obstacles.
Video demonstration of: Model-based versus Model-free Deep Reinforcement Learning for Autonomous Racing Cars
2021-03-03
oai:zenodo.org:4577412
Audiovisuelles Material
Elektronische Ressource
Englisch
DDC: | 629 |
Engineering Index Backfile | 1923
|Racing Ahead with Autonomous Cars and Digital Innovation
Springer Verlag | 2015
|TIBKAT | 1978
|