Path planning of virtual human by using reinforcement learning

Virtual Human can hold the possibility of performing a variety of assistive and analysis tasks in 3D virtual environments. However, widespread use of avatars assistants in these environments requires ease of use by individuals who are generally not skilled on designing operators. In this paper we present a method of training virtual human that bridges the gap between designing and building of a virtual humanpsilas action as well as its autonomous learning ability to adapt a certain task. With our approach, we integrate reinforcement learning to the action of virtual humanpsilas path planning to achieve fast policy acquisition. The result of the approach is illustrated in the case of successfully instructing virtual human to cross a terrain with pyramid and a number of obstacles to reach a certain target.