A mobile robot has been constructed, and a motion planning methodology for operating in an environment with unknown arbitrary obstacles has been implemented and tested in a complex cluttered environment. The authors address the following implementation issues: the robot's ability to cope with limited information available from typical sensors while operating in a cluttered environment with unknown obstacles; handling the registration problem, to assess the robot's current location; performing necessary data processing and planning fast enough to assure acceptable motion speeds; handling situations which require the robot to make loops, retreat, and visit some areas of its path more than once. The results indicated feasibility of practical systems capable of operating in a complex uncertain environment.>

Recent research on sensor-based motion planning has produced theory and algorithms for motion planning for robots operating in an environment with unknown obstacles. The implementation of such systems requires constructing a robot, sensors, and a complex control computer environment - a time-consuming and costly undertaking. To provide a necessary bridge between theory and implementation - to test algorithmic solutions and to cut down on the number of experimental alternatives - a computer simulation system for modeling sensor-based motion planning has been developed. The system provides means for modeling various robots, sensor systems, and sets of obstacles in the robot environment, testing a variety of two- and threedimensional motion planning algorithms, and testing human performance in motion planning. It also provides a real-time animation of the robot motion. The simulation is realistic in that only information from the assumed sensor media is used as input to a motion planning algorithm. These features make it a unique system for simulating robot planning and control in an uncertain environment.

The objective of the hospital transport mobile robot, HelpMate, is to carry out such tasks as the delivery of off-schedule meal trays, lab and pharmacy supplies, and patient records. Unlike many existing delivery systems in the industry which operate within a rigid network of wires buried or attached to the floor, a hospital transport robot is expected to be able to navigate much like a human, including handling uncertainty and unexpected obstacles. The navigation system of HelpMate makes use of the specific structure of a hospital hallway environment and relies on recent results on provable sensor-based motion planning algorithms that specifically address the issue of navigation in an unknown and unstructured environment. also incorporated in the system are algorithms for handling unmodeled factors such as sensor noise and sensor inaccuracy, errors in position estimation, and moving obstacles.>

A model of mobile robot navigation is considered in which the robot is a point automaton operating in an environment with unknown obstacles of arbitrary shapes. The robot's input information includes its own and the target-points coordinates as well as local sensing information such as that from stereo vision or a range finder. These algorithmic issues are addressed: (1) Is it possible to combine sensing and planning functions to produce 'active sensing' guided by the needs of planning? (The answer is yes). (2) Can richer sensing (e.g., stereo vision versus tactile) guarantee better performance, that is, resulting in shorter paths? (The general answer is no). A paradigm for combining range data with motion planning is presented. It turns out that extensive modifications of simpler tactile algorithms are needed to take full advantage of additional sensing capabilities. Two algorithms that guarantee convergence and exhibit different styles of behavior are described, and their performance is demonstrated in simulated examples.>

The authors present a paradigm for combining vision with motion planning. It turns out that extensive modifications of so-called tactile algorithms are needed to take full advantage of the additional sensing capabilities, while not sacrificing the algorithm convergence. Different design principles can be introduced that result in algorithm versions exhibiting different styles of behavior and producing different paths, without, in general, being superior to each other.>