Masked Humanoid Controller Enables Multi-Modal Whole-Body Control for Real-World Robots
There's an exciting breakthrough in humanoid robotics with the launch of the Masked Humanoid Controller (MHC). This advanced whole-body controller brings together different command interfaces. It allows high-level systems to set masked target paths using selected robot state variables. The MHC can handle various inputs, like optimized paths, motion capture, re-targeted videos, and joystick commands in real-time. It was trained through a simulation that covered these different aspects, which helps it perform well even with partially defined actions while maintaining balance and handling disturbances. They demonstrated its capabilities both in simulation and on the Digit humanoid robot, addressing a major challenge in creating a unified interface for comprehensive movements, from precise foot placements to joystick teleoperation.
Key facts
- The Masked Humanoid Controller (MHC) is a learned whole-body controller for humanoid robots.
- MHC exposes an interface based on masked target trajectories over selected subsets of the robot's state variables.
- It accommodates multi-modal inputs: optimized trajectories, motion capture clips, re-targeted video, and real-time joystick signals.
- MHC is trained in simulation using a curriculum that spans the full range of modalities.
- It enables robust execution of partially specified behaviors while maintaining balance and disturbance rejection.
- Demonstrated both in simulation and on the real-world Digit humanoid robot.
- Addresses the challenge of a unified interface for commanding diverse whole-body behaviors.
- Behaviors include precise footstep sequences, partial-body mimicry, and joystick teleoperation.
Entities
—