The problem with DDPG: understanding failures in deterministic environments with sparse rewards

Abstract

In environments with continuous state and action spaces, state-of-the-art actor-critic reinforcement learning algorithms can solve very complex problems, yet can also fail in environments that seem trivial, but the reason for such failures is still poorly understood. In this paper, we contribute a formal explanation of these failures in the particular case of sparse reward and deterministic environments. First, using a very elementary control problem, we illustrate that the learning process can get stuck into a fixed point corresponding to a poor solution. Then, generalizing from the studied example, we provide a detailed analysis of the underlying mechanisms which results in a new understanding of one of the convergence regimes of these algorithms. The resulting perspective casts a new light on already existing solutions to the issues we have highlighted, and suggests other potential approaches.

arXiv comment19 pages, submitted to ICLR 2020

References in corpus (7)

• Continuous control with deep reinforcement learning
• Playing Atari with Deep Reinforcement Learning
• Noisy Networks for Exploration
• Parameter Space Noise for Exploration
• Off-Policy Deep Reinforcement Learning without Exploration
• GEP-PG: Decoupling Exploration and Exploitation in Deep Reinforcement Learning Algorithms
• Towards Characterizing Divergence in Deep Q-Learning

Cited by in corpus (1)

• Learning Energy-Efficient Hardware Configurations for Massive MIMO Beamforming