Navigation for an Intelligent Mobile Robot - IEEE Xplore

CROWLEY: NAVIGATION FOR AN INTELLIGENT ROBOT 39represent the free space in which the robot may travel. Thespace inside a doorway after shrinking forms a special regioncalled a “doorway region.” Doorway regions are not guaranteedto be convex. Each adit is connected to the adit on theother side of the doorway and to all other adits within itsconvex region, as shown in Fig. 10. In this figure, the adits tothe convex regions are illustrated with boxes.The network of places is a three level structure. At the toplevel are a list of user-defined “named places.” At the middleis a list of convex regions. Each named place points to aconvex region, and each convex region contains a list ofnamed places. Each convex region also contains a list of adits.The adits serve as landmarks to global path planning andexecution. The convex regions serve as “legal highways” forplanning paths to any named place.C. Global Path PlanningA global path is planned as a sequence of adits which willtake the robot from its current convex region to the convexregion which contains a specified goal point. A command ofthe form “go to (place)” provides a pointer to a convex regionthat in turn provides a list of possible goal adits. The aditclosest to the named place is chosen as a goal for pathplanning. Knowledge of the current convex region gives a listof adits from which to start the path. The nearest adit isselected as a starting adit. The shortest path through thenetwork of adits is determined using a version of Dijkstra’salgorithm [ 11 which halts when a path to the desired goal placehas been found. If the start (or the end) of this path leadsthrough two adits in the same region, the first (or last) adit isdropped from the path. Global path execution is then reducedto a three step process in which the IMP 1) moves to the firstadit in the path, 2) moves from each adit on the path to thenext, and 3) then moves from the last adit to the goal place.VIII. .LOCAL NAVIGATIONFig. 11. State transition diagram for local navigation.IMPFig. 12. Legal highway for pathGoalexecution.In the TURN state, the IMP turns toward the goal point untilthe difference between the estimated orientation and thedirection to the goal point falls below the minimum turningresolution. The IMP then enters the WAIT state to make acomplete sonar scan and verify the current estimated position.The sonar scan \results in an update in the estimated positionand orientation, even if there is no change in the estimatedposition or orientation. The call to the function “Set-Estimated-Position” signals the completion of the scan andcauses a transition back to the DECIDE state. If the IMP is not atthe goal, and is turned toward the goal, control will pass fromDECIDE t0 MOVE.Upon entering the MOVE state, the IMP computes theequation of the line (the path equation) from the currentlocation to the goal point. A cyclic process is then initiated inwhich the system moves forward while performing thefollowing tests as rapidly as possible.1) Verify that the distance to the goal point is decreasing.When the distance to the goal stops decreasing, the systemreturns to HOLD to wait for the next goal. If the distance is everA local straight line path is executed by a finite state processwhich monitors the position of the robot to assure that it increasing and is larger than the goal tolerance, then theremains on the desired straight line within a tolerance. This system will go into the WAIT state to take a clean view of theprocess also monitors the local model to assure that no world.unexpected obstacle blocks the path. A recursive obstacle 2) Verify that the perpendicular distance from the currentavoidance algorithm is used to plan a path around unexpected estimated position to the path line segment is below aobstacles.A. Local Path ExecutionStraight line movement to a goal point is monitored by arelatively simple finite state process. The states of this processare the set {HOLD, DECIDE, TURN, MOVE, WAIT, BLOCKED). Thetolerance. This test is illustrated by Fig. 12. It is performed byevaluating the path equation using the current estimatedposition, yielding the perpendicular distance to the pathequation. If this distance exceeds the tolerance, then the IMPwill go to WAIT.3) Verify that there is a free path to the goal. This is donestate transition diagram for this process is shown in Fig. 11. by projecting parallel line segmentsin the composite localThe process waits in the HOLD state for a goal from the global model. If the path becomes blocked, the IMP will go intopath execution process. When a goal is received, the IMP blocked state and signal for local path planning to avoid theenters the DECIDE state. In the DECIDE state it first tests thedistance to the goal. If this distance is less than a tolerance, itreturns to HOLD. If the distance is above the tolerance, theobstacle.If the IMP is in the MOVE or TURN States and its contactsensor is triggered, it immediately halts and enters the BLOCKEDdifference in angle between the current heading and the goal is state. Entering BLOCKED triggers the local path planningtested. If this angle is above the minimum resolution for procedures to plan a path around an obstacle. A lowturning,the IMP enters the TURN state; otherwise it enters theMOVE Sbb.confidence line segment is also placed into the composite localmodel to represent the obstacle.