FastTD3: Simple, Fast, and Capable Reinforcement Learning for Humanoid Control

TL;DR

FastTD3 is a streamlined reinforcement learning algorithm that drastically reduces training time for humanoid robots, achieving high performance in under 3 hours using simple modifications and efficient training techniques.

Contribution

We introduce FastTD3, a novel RL algorithm that combines simplicity and speed, enabling rapid training of humanoid control policies with minimal computational resources.

Findings

Solves HumanoidBench tasks in under 3 hours on a single GPU

Maintains training stability with simple modifications

Provides an accessible implementation for robotics RL research

Abstract

Reinforcement learning (RL) has driven significant progress in robotics, but its complexity and long training times remain major bottlenecks. In this report, we introduce FastTD3, a simple, fast, and capable RL algorithm that significantly speeds up training for humanoid robots in popular suites such as HumanoidBench, IsaacLab, and MuJoCo Playground. Our recipe is remarkably simple: we train an off-policy TD3 agent with several modifications -- parallel simulation, large-batch updates, a distributional critic, and carefully tuned hyperparameters. FastTD3 solves a range of HumanoidBench tasks in under 3 hours on a single A100 GPU, while remaining stable during training. We also provide a lightweight and easy-to-use implementation of FastTD3 to accelerate RL research in robotics.