This paper summarizes work performed at NRaD during FY94 on the integration of robotic sensor and motor systems. Two robotic applications were involved. These are the visual control of a five degrees of freedom manipulator arm in three dimensions, and the visual control of a mobile platform for target acquisition, tracking and trailing with obstacle avoidance. Solutions to both applications were constrained by the intentional restriction that information could only be gained autonomously by the robot through vision and proprioception (internal information on mechanical position). Furthermore, the only visual information allowed was the frame-to-frame image flow derived from the contrast gradients. Both reflexive (reactive) and adaptive (learning) algorithms were studied.
展开▼