Development of an intelligent robotic aid to the physically handicapped

The design and implementation of an intelligent robotic system to assist physically handicapped people are presented. The system uses artificial-intelligence-based task planning and vision sensors. Emphasis is on flexibility and a model-based approach. The vision system has been designed to be modular and is built in three levels: low-level image acquisition, intermediate-level segmentation and feature extraction, and high-level object recognition. Once the object is recognized, its state (position and orientation) and other relevant features are transferred to the task planner. The task planner works in a hierarchy of two abstraction spaces by first generating a rough plan devoid of all unnecessary details from a user-defined goal and then refining and producing a sequence of robot-level commands. The advantage of the approach is that by first ignoring detail planning, it is possible to effectively reduce the number of subgoals to be accomplished in any given abstraction space, thus greatly reducing computation. The system was implemented on a unique robotic arm called the Soft Arm. The results of two example runs, PICKUP SPOON and FEED ME SOUP, are also shown.<>