Projects
Selected robotics projects spanning Physical AI, vision-language-action learning, bimanual and humanoid manipulation, model predictive control, sim-to-real transfer, and robot foundation models.
VLM-RAG-guided whole-body MPC framework with virtual impedance for bimanual mobile manipulation. Links semantic reasoning to compliant physical interaction, retrieving task-relevant episodes to configure impedance gains, velocity limits, and safety margins.
Isaac Sim digital twin and synthetic data-generation pipeline for tactile-aware vision-language-action learning. Enables contact-rich manipulation of fragile and deformable objects without tactile sensing at inference time.
VLM-RAG-driven impedance-control system for humanoid robots, connecting retrieved task knowledge to compliant, safe physical interaction. Published at ACM/IEEE HRI 2026.
Real-time Cartesian MPC for anthropomorphic bimanual manipulation on the Unitree H1 humanoid, with control latency under 10 ms, hardware-in-the-loop testing, and sim-to-real calibration.
Isaac Lab reinforcement-learning pipeline for Unitree H1 motion imitation, using custom rewards and curriculum learning to reproduce expressive whole-body dance motions.
AnyGoal explores vision-language-guided multi-agent exploration for training-free lifelong navigation, connecting semantic goal understanding with scalable autonomous exploration.