Project
Ongoing Project
Improve the Sim-to-Real Transfer for Control of Robots via the Koopman Operator Theory
We aim to achieve "smoother" sim-to-real transfer by embedding the Koopman operator.
Past Project
Data-driven Hierarchical Control Structure with Koopman Operator Theory
Design of a hierarchical control structure with Koopman operator theory to improve the performance of robots under uncertainty.
The structure is evaluated and justified in the Crazyflie micro-aerial robots that are wrapped around the Geometric controller and PID controller.
Online Modeling and Control of Soft Robots using Koopman Operator
We design a model predictive control structure with the online updated model constraints using the Koopman operator to drive soft grippers.
The gripper could grasp different objects with varying weights and shapes.
Design of the Lifting Functions in Koopman Operator Estimation
We propose a general and analytical methodology to formalize the construction of lifting functions based on system's characteristic properties of a robot.
We evaluate the structure with a range of diverse nonlinear robotic systems (a wheeled mobile robot, a two-revolute-joint robotic arm, and a soft robotic leg).
Analysis on robustness to noise of model extraction and prediction using Koopman Operator
We build the theoretical contribution on model extraction and states prediction that proposes a way to quantify the prediction error because of noisy measurements when using Koopman operator to estimate the system.
Then a non-holonomic wheeled robot, ROSbot, is simulated in Gazebo to illustrated the performance of the algorithm.