Advanced robotics
The primary challenges in advanced robotics are associated with adaptation and autonomy. Mechatronic systems such as drones, mobile robots and manipulators need to be able to adapt, self-learn and auto-reconfigure to changes in tasks, as well as operate in unstructured environments and handle uncertainty, without the need for skilled, systems configuration personnel.
Our research is on enabling machines to learn without being explicitly programmed, so that they can progressively improve their performance on a specific task using data only. Scalability is achieved by high-level machine teaching via human-machine communications at the semantic level. By employing inverse reinforcement learning approaches, we address imitation learning whereby a robot can learn by watching a human operator. The model-based approach that we follow establishes a common framework for effective human-machine communication throughout the process of learning and deployment, and efficient control of design and implementation.