Deep Policy Gradient Methods Without Batch Updates, Target Networks, or Replay Buffers

TL;DR

This paper introduces Action Value Gradient (AVG), a novel incremental deep policy gradient method that enables effective reinforcement learning on robots without large replay buffers or batch updates, addressing resource limitations.

Contribution

The paper proposes AVG, the first effective incremental deep policy gradient method, along with normalization techniques, allowing reinforcement learning on real robots with limited computational resources.

Findings

AVG performs comparably to batch methods on simulation benchmarks.

AVG enables reinforcement learning on real robots with incremental updates.

The method addresses instability issues in incremental deep learning.

Abstract

Modern deep policy gradient methods achieve effective performance on simulated robotic tasks, but they all require large replay buffers or expensive batch updates, or both, making them incompatible for real systems with resource-limited computers. We show that these methods fail catastrophically when limited to small replay buffers or during incremental learning, where updates only use the most recent sample without batch updates or a replay buffer. We propose a novel incremental deep policy gradient method -- Action Value Gradient (AVG) and a set of normalization and scaling techniques to address the challenges of instability in incremental learning. On robotic simulation benchmarks, we show that AVG is the only incremental method that learns effectively, often achieving final performance comparable to batch policy gradient methods. This advancement enabled us to show for the first time effective deep reinforcement learning with real robots using only incremental updates, employing a robotic manipulator and a mobile robot.