Reinforcement Learning and Dynamic Planning Applied to Virtual Humans Animation

We present an approach using learning and planning techniques to deal with the problem of animating virtual humans in 3D environments. The main idea is to use learning to guide the initial behavior for animation, then, if an unexpected situation appears use the dynamic planning algorithm to go one step more to finish the animation. The advantage of this strategy is to improve the time required by algorithms to animate virtual creatures. Results obtained show how using this strategy improves the performance regarding just using the planning algorithms.