We aim to develop a systematic framework for robots to build models of the world and to use these to make effective and safe choices of actions to take in complex scenarios.
Imagine a robot butler operating in a new home. Unlike factory robots that can follow almost the same trajectories to complete the same task repeatedly, the robot butler needs to adapt itself to new environments and learn how to act for a variety of tasks. In addition, the home environment is more uncertain than a factory and the robot butler needs to understand how things work so as to get things done. Our goal in this project is to develop a systematic framework for such robots to learn to understand how things work in their surroundings and make effective and safe choices of actions to take. We study the fundamentals of model learning (understanding the world) from actively selected data and planning (making sequential decisions to complete a task) with the help of optimization techniques (e.g. Bayesian Optimization for optimizing expensive functions, Stochastic Gradient Descent for function fitting), develop new algorithms with theoretical guarantees, and test in realistic scenarios.