The authors present an algorithm designed to provide a robot-vehicle with sufficient 'intelligence' to be able to optimize its behavior in the light of information while carrying out the task of moving from any position and orientation to any other position and orientation. The kinematic and dynamic constraints of a nonholonomic vehicle are included explicitly. It is thus possible to finesse issues concerned with producing planned paths that cannot be achieved in practice by a real vehicle. Some of the maneuvers that this algorithm can support are shown. The algorithm has been designed to allow the inclusion of unmapped obstacles and thus has the potential of being extended to situations where there are multiple moving vehicles.>

A homing and obstacle avoiding algorithm is described. It takes as its input the current position and heading of a robot-vehicle relative to a desired position and heading and then 'maps' the 'difference' between them directly into actions for the controllers of motion and heading to act on. It includes explicitly, and in a separate phase of the overall computation the kinematic and dynamic constraints of a non-holonomic vehicle. The algorithm has been designed to allow range and bearing information about unmapped obstacles to modify the actions of a robot-vehicle. The authors show some of the obstacle avoiding behaviour. Since the objects are unmapped they could be other moving vehicles. To overcome uncertainty in the actual movement the algorithm continually recomputes the control actions needed to achieve the goal. Thus the algorithm has the potential to be extended to situations where the goal location is changing.>

The navigation and guidance of a mobile robot depend on inferences made from sensor measurements. Acoustic rangers offer the prospect of a simple means of determining range and direction but cannot in general be relied upon to measure the true range. To overcome this problem acoustic cats' eyes (the acoustical analog of optical retroreflectors found on highways) were attached to the smooth vertical surfaces in the laboratory. A standard Polaroid sonar ranger was rotated through one complete circle while range measurements were taken at regular intervals. This experiment was repeated without these devices. The range readings were then superimposed onto an architect's plan view of the area. The range scans taken with the acoustic cats' eyes show a significantly higher degree of correspondence to the plan view than do the range scans taken without them. It is shown that by identifying step changes in the derivative (the curvature) with respect to direction a decision space is created from which the presence and location of corners, walls, and doorways can be inferred.>