Event

DeepMind trains agile soccer skills on ROBOTIS OP3 with deep reinforcement learning

Apr 26, 2023 · Research Publication · ROBOTIS · Actuation

Key points

  • The DeepMind paper trains a 20-actuator miniature humanoid robot to play one-versus-one soccer using deep reinforcement learning and self-play.
  • The learned policy transfers zero-shot from simulation to real robots and improves walking speed, turning speed, get-up time, and kicking speed versus scripted baselines.
  • The result gives OP3-W a high-profile benchmark for sim-to-real reinforcement learning on agile humanoid soccer skills.

Company context

Develops actuator systems, control platforms, and modular robotics kits, best known for its DYNAMIXEL smart servos used across research, education, and commercial robotics. The company also builds complete robotic platforms including humanoid and mobile robots, primarily as development and ecosystem tools.

Context

Company
ROBOTIS
Segment
Actuation
Event type
Research Publication
Geography
Seoul · South Korea

Links