Learning to drive from a world on rails
This work addresses the problem of sample-efficient autonomous driving for researchers and practitioners, though it is incremental as it builds on model-based approaches with a simplifying assumption.
The paper tackles learning a vision-based driving policy from pre-recorded logs by assuming a non-reactive world, which simplifies dynamics and enables efficient training. It achieves state-of-the-art results, ranking first on the CARLA leaderboard with a 25% higher driving score using 40 times less data and showing high sample efficiency on ProcGen benchmarks.
We learn an interactive vision-based driving policy from pre-recorded driving logs via a model-based approach. A forward model of the world supervises a driving policy that predicts the outcome of any potential driving trajectory. To support learning from pre-recorded logs, we assume that the world is on rails, meaning neither the agent nor its actions influence the environment. This assumption greatly simplifies the learning problem, factorizing the dynamics into a nonreactive world model and a low-dimensional and compact forward model of the ego-vehicle. Our approach computes action-values for each training trajectory using a tabular dynamic-programming evaluation of the Bellman equations; these action-values in turn supervise the final vision-based driving policy. Despite the world-on-rails assumption, the final driving policy acts well in a dynamic and reactive world. At the time of writing, our method ranks first on the CARLA leaderboard, attaining a 25% higher driving score while using 40 times less data. Our method is also an order of magnitude more sample-efficient than state-of-the-art model-free reinforcement learning techniques on navigational tasks in the ProcGen benchmark.