Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Berkeley Humanoid: A Research Platform for Learning-based Control

Qiayuan Liao, Bike Zhang, Xuanyu Huang, Xiaoyu Huang, Zhongyu Li, Koushil Sreenath

TL;DR

Berkeley Humanoid presents a low-cost, mid-scale humanoid platform designed to enable fast, learning-based locomotion with a narrow sim-to-real gap. By combining a compact 16 kg robot, direct joint actuation via quasi-direct-drive actuators, hollow shafts, and high-bandwidth EtherCAT control, the authors minimize modeling and latency errors while enabling accurate domain randomization guided by hardware-informed parameters. A minimally composed PPO-based policy, operated through simple PD control of target joint positions, demonstrates robust omnidirectional walking, terrain adaptation, disturbance rejection, and long-distance outdoor locomotion, including hopping. The work demonstrates that careful hardware design, together with simulation-aware control, can yield agile, reliable learning-based locomotion on outdoor terrains with a small sim-to-real gap, enabling scalable real-world deployment.

Abstract

We introduce Berkeley Humanoid, a reliable and low-cost mid-scale humanoid research platform for learning-based control. Our lightweight, in-house-built robot is designed specifically for learning algorithms with low simulation complexity, anthropomorphic motion, and high reliability against falls. The robot's narrow sim-to-real gap enables agile and robust locomotion across various terrains in outdoor environments, achieved with a simple reinforcement learning controller using light domain randomization. Furthermore, we demonstrate the robot traversing for hundreds of meters, walking on a steep unpaved trail, and hopping with single and double legs as a testimony to its high performance in dynamical walking. Capable of omnidirectional locomotion and withstanding large perturbations with a compact setup, our system aims for scalable, sim-to-real deployment of learning-based humanoid systems. Please check http://berkeley-humanoid.com for more details.

Berkeley Humanoid: A Research Platform for Learning-based Control

TL;DR

Berkeley Humanoid presents a low-cost, mid-scale humanoid platform designed to enable fast, learning-based locomotion with a narrow sim-to-real gap. By combining a compact 16 kg robot, direct joint actuation via quasi-direct-drive actuators, hollow shafts, and high-bandwidth EtherCAT control, the authors minimize modeling and latency errors while enabling accurate domain randomization guided by hardware-informed parameters. A minimally composed PPO-based policy, operated through simple PD control of target joint positions, demonstrates robust omnidirectional walking, terrain adaptation, disturbance rejection, and long-distance outdoor locomotion, including hopping. The work demonstrates that careful hardware design, together with simulation-aware control, can yield agile, reliable learning-based locomotion on outdoor terrains with a small sim-to-real gap, enabling scalable real-world deployment.

Abstract

We introduce Berkeley Humanoid, a reliable and low-cost mid-scale humanoid research platform for learning-based control. Our lightweight, in-house-built robot is designed specifically for learning algorithms with low simulation complexity, anthropomorphic motion, and high reliability against falls. The robot's narrow sim-to-real gap enables agile and robust locomotion across various terrains in outdoor environments, achieved with a simple reinforcement learning controller using light domain randomization. Furthermore, we demonstrate the robot traversing for hundreds of meters, walking on a steep unpaved trail, and hopping with single and double legs as a testimony to its high performance in dynamical walking. Capable of omnidirectional locomotion and withstanding large perturbations with a compact setup, our system aims for scalable, sim-to-real deployment of learning-based humanoid systems. Please check http://berkeley-humanoid.com for more details.
Paper Structure (39 sections, 10 figures, 6 tables)

This paper contains 39 sections, 10 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Humanoid Design.
  4. Humanoid Control.
  5. Design for Learning-based Control
  6. System Overview
  7. Simulation-Friendly
  8. Motivation.
  9. Our Approach.
  10. Reliable and Low-Cost
  11. Motivation.
  12. Our Approach.
  13. Experiment-Friendly
  14. Motivation.
  15. Our Approach.
  16. ...and 24 more sections

Figures (10)

  • Figure 1: Design, training, and sim-to-real deployment of our custom-built humanoid with a learning-based controller.
  • Figure 2: Overview of design: (a) main components, (b) joints and key dimensions, (c) key actuators and joints of the left leg.
  • Figure 3: (a) Exposed view and (b) cross view of one of our custom actuators.
  • Figure 4: Omnidirectional Walking. (a-c) The robot walks forward, turns in place, and walks backward in the lab environment. (d, e) The robot walks forward and sideways in the wild.
  • Figure 5: Walking on Various Terrains. (a) The robot walks on eight different types of terrain. (b) The robot climbs a relatively steep and narrow unpaved trail covered with dust and rocks. (c) The robot walks on an uneven pathway. (d) The robot makes a turn on rocky stairs.
  • ...and 5 more figures