Exploring Model-based Planning with Policy Networks
Tingwu Wang, Jimmy Ba

TL;DR
This paper introduces POPLIN, a model-based reinforcement learning algorithm that combines policy networks with online planning, achieving state-of-the-art sample efficiency in complex environments by optimizing action sequences and policy parameters.
Contribution
The paper proposes a novel algorithm, POPLIN, integrating policy networks with online planning, and demonstrates its superior performance and smoother optimization surface compared to existing methods.
Findings
POPLIN achieves about 3x more sample efficiency than PETS, TD3, and SAC.
Optimization in parameter space results in a smoother surface, improving planning.
Distilled policy networks can be used without model predictive control during testing.
Abstract
Model-based reinforcement learning (MBRL) with model-predictive control or online planning has shown great potential for locomotion control tasks in terms of both sample efficiency and asymptotic performance. Despite their initial successes, the existing planning methods search from candidate sequences randomly generated in the action space, which is inefficient in complex high-dimensional environments. In this paper, we propose a novel MBRL algorithm, model-based policy planning (POPLIN), that combines policy networks with online planning. More specifically, we formulate action planning at each time-step as an optimization problem using neural networks. We experiment with both optimization w.r.t. the action sequences initialized from the policy network, and also online optimization directly w.r.t. the parameters of the policy network. We show that POPLIN obtains state-of-the-art…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsReinforcement Learning in Robotics · Robotic Locomotion and Control · Real-time simulation and control systems
MethodsExperience Replay · Dense Connections · *Communicated@Fast*How Do I Communicate to Expedia? · Target Policy Smoothing · Clipped Double Q-learning · Adam · Twin Delayed Deep Deterministic
