Learning Latent Dynamics for Planning from Pixels

TL;DR

The paper introduces PlaNet, a model-based agent that learns environment dynamics from pixel inputs and plans efficiently in latent space, enabling it to solve complex control tasks with fewer episodes than traditional methods.

Contribution

It presents a novel latent dynamics model with stochastic components and a multi-step variational inference objective called latent overshooting, improving planning accuracy from images.

Findings

PlaNet outperforms previous methods on complex control tasks.

It achieves high performance with fewer episodes than model-free algorithms.

The approach effectively handles partial observability and sparse rewards.

Abstract

Planning has been very successful for control tasks with known environment dynamics. To leverage planning in unknown environments, the agent needs to learn the dynamics from interactions with the world. However, learning dynamics models that are accurate enough for planning has been a long-standing challenge, especially in image-based domains. We propose the Deep Planning Network (PlaNet), a purely model-based agent that learns the environment dynamics from images and chooses actions through fast online planning in latent space. To achieve high performance, the dynamics model must accurately predict the rewards ahead for multiple time steps. We approach this using a latent dynamics model with both deterministic and stochastic transition components. Moreover, we propose a multi-step variational inference objective that we name latent overshooting. Using only pixel observations, our agent solves continuous control tasks with contact dynamics, partial observability, and sparse rewards, which exceed the difficulty of tasks that were previously solved by planning with learned models. PlaNet uses substantially fewer episodes and reaches final performance close to and sometimes higher than strong model-free algorithms.