GMO Builds Sprinting Humanoid from Athlete Mocap

🔑 Enhanced Key Takeaways

• •The project leverages GMO Internet Group's existing 'GMO Athletes' professional track and field team, utilizing high-fidelity biomechanical data captured from elite sprinters to train the humanoid's gait and balance controllers.
• •The initiative is part of a broader strategic pivot by GMO Internet Group to integrate AI-driven robotics into their 'Internet Infrastructure' and 'Online Advertising' business segments, aiming to create new revenue streams in physical automation.
• •The development focuses on overcoming the 'actuator density' challenge, requiring custom high-torque, high-speed motor designs that can withstand the extreme impact forces generated during bipedal sprinting, which differ significantly from standard walking-gait humanoid designs.

🛠️ Technical Deep Dive

• •Architecture: Employs a hierarchical control system where a high-level policy (trained via Reinforcement Learning) maps Mocap data to low-level joint torque commands.
• •Actuation: Utilizes proprietary high-bandwidth, quasi-direct drive actuators designed to minimize impedance and maximize power-to-weight ratios for explosive movement.
• •Sensor Fusion: Integrates high-frequency IMU data with visual-inertial odometry to maintain balance during high-velocity transitions and rapid directional changes.
• •Simulation: Uses a custom physics engine environment that mimics the specific friction coefficients and surface elasticity of professional track materials (e.g., synthetic rubber tracks).

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline