Research Interests
I am interested in integrating learning and control to build robots that work reliably in the real world. Learned policies offer flexibility and generalization, while feedback control offers precision, robustness, and guarantees during physical interaction. I bring the two together by building structure from geometry, dynamics, and physical interaction into both, so that the whole system, from task-level policy to low-level controller, shares the same structure.
Current Research
- Geometric impedance control (GIC) on SE(3) — unified position and orientation control; Lie group- vs. Lie algebra-based potential functions; unified force-impedance control (GUFIC) with passivity guarantees
- SE(3)-equivariant robot learning and control — equivariant control law for contact-rich manipulation; EquiContact, a vision-to-force equivariant policy that generalizes to unseen task poses; tutorial survey on equivariant learning
- Precise humanoid manipulation — geometric admittance control (GAC) with residual RL (ResGAC) for sub-centimeter end-effector tracking in whole-body humanoid control
Earlier Work
- Nonlinear control for UAVs — approximate dynamic inversion & extended high-gain observers; explicit nonlinear MPC; recursive least-squares-based dynamic inversion
- Deep reinforcement learning for a semi-active suspension system
- Gain-scheduling robust control (LPV/LMI) — road-adaptive semi-active suspension with an LSTM-based road parameter estimator; autonomous vehicle control under vision uncertainties
- Safety-critical control with CBFs — adaptive cruise control based on a relaxed control barrier function and dynamic inversion