Industrial use of automated guided vehicles (AGVS) requires execution of mo
bile tasks in populated work areas. How these robots can arrive at their go
als without collision with potential obstacles and how the knowledge acquir
ed about obstacles in earlier movements may be used in later navigation are
the key issues in this paper. A collision avoidance algorithm integrating
use of proximity sensors is developed which can be adapted to any two-dimen
sional work area. The algorithm also involves construction and continual up
date of an obstacle matrix using information acquired during earlier moveme
nts. Once the obstacle matrix for the work area is identified, following na
vigation can be executed with ease using an off-line collision avoidance al
gorithm. The algorithm developed here is an attempt to combine the merits o
f both on-line and off-line collision avoidance strategies. Before commandi
ng the AGV to move, the algorithm will compute an off-line collision-free t
rajectory based on current knowledge regarding obstacles in the work area.
Since knowledge of the current rate of obstacle topology may be incomplete,
it is entirely possible that the AGV may detect a potential candidate for
collision while travelling along its precomputed trajectory. Should such a
situation occur, the AGV will immediately invoke an on-line strategy which
will enable the AGV to avoid collision. Upon successfully avoiding the imme
diate candidate for collision, the AGV will again recompute an off-line pat
h and the process will continue. A stationary robot is used to test the fea
sibility of the on-line component of the algorithm by guiding its end effec
tor through an obstacle maze. The off-line algorithm Jim has been simulated
on a 80386-based personal computer. Examples are included to demonstrate t
he capability of the algorithm.