Version 1.0, August 31, 2001, Copyright, Hugh Jack 1993-2001
The amount of knowledge which a path planner has may be very limited. If the robot has no previous knowledge of the environment, then information must be gathered while the robot is in motion. Trajectory planners rely on feedback for finding new trajectories and detecting poor results. Contact or distance sensors are used to detect an obstacle and the manipulator trajectory is altered to avoid collision. This method will typically guarantee a solution (if it exists or if it does not encounter a blind alley), but at a much higher time cost, and a longer path. The collection of current data becomes critical when dealing with moving obstacles, that do not have a periodic cycle. This method may also be tested by simulation as suggested by K.Sun and V.Lumelsky [1987], who developed a simulator for a sensor based robots.
For the purpose of clarifying this subject a special distinction will be drawn between a path and a trajectory. When discussing a path, it will refer to the complete route traced from the start to the goal node. The path is made up of a number of segments and each of these path segments is continuous (no stop points, or sharp corners). Another name for a path segment could be a trajectory. This distinction is presented as being significant, by the author, when considering a trajectory planner, which basically chooses the locally optimum direction, as opposed to a complete path. Only some path planners use trajectory based planning, which is easier and faster to compute, but generally produces sub-optimal paths.