Robot Learning
We aim at integrating control theory with machine learning to enable intelligent robotic behavior in uncertain and dynamic environments.
Collectively, this research advances adaptive, explainable, and capable robotic systems by uniting learning and control. Research on this direction include:
- Neuro-adaptive robot control.
- Modeling physical system and processes with machine learning tools. Examples:
- Gaussian Process Regression for manipulation affordances
- Variational inference for road friction estimation
- Data-driven model predictive control (MPC) for robotic applications:
- Food cutting
- Navigation in human-centered environments through motion prediction
- Dynamic Movement Primitives:
- Prediction of human motion for human robot interaction
- Scaling or temoral coupling control
- Reinforcement Learning for Continuous Robot Control
- Deformable object shaping with simulation based and offline reinforcement learning
