Dynamics Learning and Control

Accurate models of robot dynamics are critical for motion planning, optimal control and generalization to novel operational conditions. My research looks at the problem of dynamics learning from a hybrid perspective, where prior physics knowledge, such as the law of energy conservation, is used to assist the learning process and control designs.

Port-Hamiltonian Neural ODE Networks on Lie Groups For Robot Dynamics Learning and Control

We develop a (port-)Hamiltonian neural ODE network on Lie groups that preserves the law of energy conservation and Lie group constraints in the dynamics model, and design an energy-based trajectory tracking controller for various robot platforms.

T-RO'24, ICRA'24, L-CSS'22, RSS'21

Adaptive-Frequency Model Learning and Predictive Control for Dynamic Maneuvers on Legged Robots

We propose a novel learning-based control approach that learns a residual dynamics model from experiment data, accounting for leg dynamics and mismatch in hardware, and enhances the prediction accuracy of the jumping trajectory for MPC.

R-AL'25

High Accuracy Aerial Maneuvers on Legged Robots using Variational Integrator Discretized Trajectory Optimization

We develop a whole-body trajectory optimization method that uses variational integration (VI) and full-body nonlinear dynamics for long-flight aggressive maneuvers such as backflipping with only few degrees of landing angle errors.

ICRA'25

Robust and Safe Autonomous Navigation for Systems with Learned SE(3) Hamiltonian Dynamics

This paper proposes techniques to learn the dynamics models of a mobile robot from trajectory data and synthesize a tracking controller with safety and stability guarantees via reference governor techniques.

OJ-CSYS'22, L4DC'22

Physics-Informed Multi-Agent Reinforcement Learning for Distributed Multi-Robot Problems

We propose a physics-informed reinforcement learning approach able to learn distributed multi-robot control policies that are both scalable and make use of all the available information to each robot.

T-RO'25, ICRA'23

Learned IMU Bias Prediction for Invariant Visual Inertial Odometry

We design a neural network to predict the bias of an inertial measurement unit (IMU) from a sequence of previous IMU measurements. The learned bias can be used for visual inertial odometry, even when visual information is unavailable for extended periods.

R-AL'25