From Simulation to Real World Maneuver Execution using Deep Reinforcement Learning

  • 2021-04-28 08:11:02
  • Alessandro Paolo Capasso, Giulio Bacchiani, Alberto Broggi
  • 0


Deep Reinforcement Learning has proved to be able to solve many control tasksin different fields, but the behavior of these systems is not always asexpected when deployed in real-world scenarios. This is mainly due to the lackof domain adaptation between simulated and real-world data together with theabsence of distinction between train and test datasets. In this work, weinvestigate these problems in the autonomous driving field, especially for amaneuver planning module for roundabout insertions. In particular, we present asystem based on multiple environments in which agents are trainedsimultaneously, evaluating the behavior of the model in different scenarios.Finally, we analyze techniques aimed at reducing the gap between simulated andreal-world data showing that this increased the generalization capabilities ofthe system both on unseen and real-world scenarios.


Quick Read (beta)

loading the full paper ...