Path planning for mobile robots using a video camera network

We describe a novel method to plan paths for mobile robots in an environment observed by a network of video cameras. The multiple video streams are fused to generate a spatial-temporal map of freespace. Updated at 5 Hz, this map concurrently locates both static and dynamic obstacles. Our novel method computes a polynomial-shaped path from current position to goal position for each map update (at 5 Hz), thus accomplishing both path planning and dynamic obstacle avoidance simultaneously. We demonstrate our methods operating in several dynamic scenarios, and compare our methods against paths planned using a numeric potential field.