Towards Generalization and Simplicity in Continuous Control

TL;DR

This paper demonstrates that simple linear and RBF policies can effectively solve continuous control tasks, offering competitive performance and better generalization compared to complex neural network policies, especially in diverse and perturbation scenarios.

Contribution

It shows that simple policy parameterizations can match complex models in continuous control, emphasizing the importance of diverse training for improved generalization.

Findings

Simple policies perform competitively on benchmarks.

Diverse initial states improve policy generalization.

Global policies recover from large perturbations.

Abstract

This work shows that policies with simple linear and RBF parameterizations can be trained to solve a variety of continuous control tasks, including the OpenAI gym benchmarks. The performance of these trained policies are competitive with state of the art results, obtained with more elaborate parameterizations such as fully connected neural networks. Furthermore, existing training and testing scenarios are shown to be very limited and prone to over-fitting, thus giving rise to only trajectory-centric policies. Training with a diverse initial state distribution is shown to produce more global policies with better generalization. This allows for interactive control scenarios where the system recovers from large on-line perturbations; as shown in the supplementary video.