Prediction of Rapidly Changing Environmental Dynamics for Real Time Behavior Adaptation using Visual Information

This paper features a method for acting in a real world environment with rapid dynamics, based on behaviors with different complexity, that emerge from: (1) direct sensing (2) on-line prediction about the future development of the environmental dynamics, and (3) internal restrictions derived by the robots strategy. The method is built upon the understanding of perception as dynamic integration of sensing, expectations, and behavioral goals, which is necessary when the environmental dynamics depends also on other intelligent agents. Two central aspects are considered: prediction of the development of environmental dynamics and the subsequent integration. The integration captures the dynamics of processes that happen in different temporal intervals but relate to the same perception. A real time object tracking method that is used is briefly described. Experiments made in a RoboCup environment with physical robots illustrate the plausibility of the method.