Towards Autonomous Synthetic Actors

This paper describes an animation approach with autonomous actors reacting to their environment and taking decisions based on perception systems, memory and reasoning. With such a system, we should be able to create simulations of situations such as virtual humans moving in a complex environment they may know and recognize, or playing ball games based on their visual and touching perception. In particular, the paper describes an animation approach where synthetic vision is used for navigation by a synthetic actor. The vision is the main channel of information between the actor and its environment and offers a universal approach to pass the necessary information from the environment to an actor in the problems of path searching, obstacle avoidance, and internal knowledge representation with learning and forgetting characteristics. For the general navigation problem, we propose a local and a global approach. In the global approach, a dynamic occupancy octree grid serves as global 3D visual memory and allows an actor to memorize the environment that he sees and to adapt it to a changing and dynamic environment. His reasoning process allows him to find 3D paths based on his visual memory by avoiding impasses and circuits. In the local approach, low level vision based navigation reflexes, normally performed by intelligent actors, are simulated. The local navigation model uses the direct input information from his visual environment to reach goals or subgoals and to avoid unexpected obstacles. A more complex example of vision-based tennis playing is also presented.