Benchmarking Deep Reinforcement Learning for Continuous Control

TLDR

Recent advances combine deep learning feature representations with reinforcement learning, enabling agents to play Atari games from pixels and acquire manipulation skills from raw sensory inputs, yet progress in continuous control remains hard to quantify without a common benchmark. We present a benchmark suite of continuous control tasks. The suite includes classic tasks such as cart‑pole swing‑up, high‑dimensional tasks like 3D humanoid locomotion, partially observed tasks, and hierarchical tasks. Systematic evaluation of various reinforcement learning algorithms on the benchmark revealed novel findings, and the benchmark along with reference implementations is publicly released to promote reproducibility and wider adoption.

Abstract

Recently, researchers have made significant progress combining the advances in deep learning for learning feature representations with reinforcement learning. Some notable examples include training agents to play Atari games based on raw pixel data and to acquire advanced manipulation skills using raw sensory inputs. However, it has been difficult to quantify progress in the domain of continuous control due to the lack of a commonly adopted benchmark. In this work, we present a benchmark suite of continuous control tasks, including classic tasks like cart-pole swing-up, tasks with very high state and action dimensionality such as 3D humanoid locomotion, tasks with partial observations, and tasks with hierarchical structure. We report novel findings based on the systematic evaluation of a range of implemented reinforcement learning algorithms. Both the benchmark and reference implementations are released at https://github.com/rllab/rllab in order to facilitate experimental reproducibility and to encourage adoption by other researchers.