A Fast Parallel Algorithm for Comparing Real and Synthetic Video

Robots are needed that can help people perform tasks in settings that are dangerous or repetitive and dull, in homes, in hospitals and in factories. This requires sophisticated human-robot interaction in which the robot can comprehend and predict human motions and plan responses that are cooperative and avoid harm. Our system uses a highly accurate physically realistic simulator coupled with stereo vision for 3D modeling and navigation that enables a robot to model and respond to human movements. The simulator is based on a 3D virtual world that is a virtual copy of the robot’s environment. A central issue is that this virtual world must be updated in real-time for it to be useful. This paper describes a fast algorithm for detecting differences between the real world and the simulated world so that the robot can update the simulated world.