Parameter Space Noise for Exploration

TL;DR

This paper introduces parameter space noise for exploration in deep reinforcement learning, demonstrating it improves learning efficiency over traditional action space noise and evolutionary strategies across various environments.

Contribution

It proposes a novel method of adding noise directly to parameters, combining benefits of exploration and sample efficiency in RL algorithms.

Findings

Parameter space noise enhances exploration effectiveness.

RL with parameter noise learns faster than traditional methods.

Applicable to both discrete and continuous control tasks.

Abstract

Deep reinforcement learning (RL) methods generally engage in exploratory behavior through noise injection in the action space. An alternative is to add noise directly to the agent's parameters, which can lead to more consistent exploration and a richer set of behaviors. Methods such as evolutionary strategies use parameter perturbations, but discard all temporal structure in the process and require significantly more samples. Combining parameter noise with traditional RL methods allows to combine the best of both worlds. We demonstrate that both off- and on-policy methods benefit from this approach through experimental comparison of DQN, DDPG, and TRPO on high-dimensional discrete action environments as well as continuous control tasks. Our results show that RL with parameter noise learns more efficiently than traditional RL with action space noise and evolutionary strategies individually.