Timezone: »
Goal-conditioned reinforcement learning endows an agent with a large variety of skills, but it often struggles to solve tasks that require more temporally extended reasoning. In this work, we propose to incorporate imagined subgoals into policy learning to facilitate learning of complex tasks. Imagined subgoals are predicted by a separate high-level policy, which is trained simultaneously with the policy and its critic. This high-level policy predicts intermediate states halfway to the goal using the value function as a reachability metric. We don’t require the policy to reach these subgoals explicitly. Instead, we use them to define a prior policy, and incorporate this prior into a KL-constrained policy iteration scheme to speed up and regularize learning. Imagined subgoals are used during policy learning, but not during test time, where we only apply the learned policy. We evaluate our approach on complex robotic navigation and manipulation tasks and show that it outperforms existing methods by a large margin.
Author Information
Elliot Chane-Sane (INRIA Paris)
Cordelia Schmid (Inria/Google)
Ivan Laptev (INRIA Paris)
Related Events (a corresponding poster, oral, or spotlight)
-
2021 Poster: Goal-Conditioned Reinforcement Learning with Imagined Subgoals »
Wed. Jul 21st 04:00 -- 06:00 AM Room
More from the Same Authors
-
2020 Poster: Radioactive data: tracing through training »
Alexandre Sablayrolles · Douze Matthijs · Cordelia Schmid · Herve Jegou -
2019 Poster: White-box vs Black-box: Bayes Optimal Strategies for Membership Inference »
Alexandre Sablayrolles · Douze Matthijs · Cordelia Schmid · Yann Ollivier · Herve Jegou -
2019 Oral: White-box vs Black-box: Bayes Optimal Strategies for Membership Inference »
Alexandre Sablayrolles · Douze Matthijs · Cordelia Schmid · Yann Ollivier · Herve Jegou