PROJECT TITLE :

Bayesian Nonparametric Reward Learning From Demonstration

ABSTRACT:

Learning from demonstration provides an attractive solution to the problem of teaching autonomous systems the way to perform complex tasks. Reward learning from demonstration is a promising methodology of inferring a rich and transferable representation of the demonstrator's intents, however current algorithms suffer from intractability and inefficiency in massive domains because of the idea that the demonstrator is maximizing one reward operate throughout the entire task. This paper takes a completely different perspective by assuming that the reward perform behind an unsegmented demonstration is actually composed of several distinct subtasks chained together. Leveraging this assumption, a Bayesian nonparametric reward-learning framework is presented that infers multiple subgoals and reward functions inside one unsegmented demonstration. The new framework is developed for discrete state areas and conjointly general continuous demonstration domains using Gaussian process reward representations. The algorithm is shown to possess each performance and computational blessings over existing inverse reinforcement learning ways. Experimental results are given in both cases, demonstrating the power to find out difficult maneuvers from demonstration on a quadrotor and a remote-controlled car.