Learning Accurate Long-term Dynamics for Model-based Reinforcement Learning

TL;DR

This paper introduces a trajectory-based model for long-term dynamics prediction in robotic systems, which improves accuracy and sample efficiency over traditional methods by directly predicting future states at specified time indices.

Contribution

The paper proposes a novel trajectory-based modeling approach that enhances long-term prediction accuracy and stability in model-based reinforcement learning.

Findings

Trajectory-based models outperform traditional models in long-term prediction accuracy.

The approach improves sample efficiency in robotic tasks.

It enables direct prediction of task rewards from the model.

Abstract

Accurately predicting the dynamics of robotic systems is crucial for model-based control and reinforcement learning. The most common way to estimate dynamics is by fitting a one-step ahead prediction model and using it to recursively propagate the predicted state distribution over long horizons. Unfortunately, this approach is known to compound even small prediction errors, making long-term predictions inaccurate. In this paper, we propose a new parametrization to supervised learning on state-action data to stably predict at longer horizons -- that we call a trajectory-based model. This trajectory-based model takes an initial state, a future time index, and control parameters as inputs, and directly predicts the state at the future time index. Experimental results in simulated and real-world robotic tasks show that trajectory-based models yield significantly more accurate long term predictions, improved sample efficiency, and the ability to predict task reward. With these improved prediction properties, we conclude with a demonstration of methods for using the trajectory-based model for control.