We build whole-body mobile manipulation for humanoid robots — enabling robust real-world behaviors that tightly couple locomotion, whole-body control, and contact-rich manipulation.

Our long-term goal is to develop learning-based systems that can move, reach, interact, and recover in complex environments, with reliable sim-to-real transfer and scalable data + training pipelines.

• Whole-body mobile manipulation

  • Coordinated locomotion + manipulation
  • Long-horizon stability and task-centric control

• Contact-rich whole-body control

  • Balancing, recovery, pushes/perturbations, and physical interaction
  • Safety-aware control and robust execution

• Dexterous manipulation

  • Bimanual skills and fine-grained control
  • Multimodal policies (vision / touch / proprioception)

• Teleoperation & data

  • Human demonstration capture and scalable dataset curation
  • Retargeting and cross-interface motion tracking

• Sim-to-real deployment

  • Domain adaptation / residual learning / system identification
  • Reproducible training, evaluation, and deployment toolchains