The problem is posed in configuration space; an algorithm (sampling-based like RRT/PRM, potential field or grid method) searches this space for a continuous path from start to goal that avoids obstacles, and the result is then smoothed and executed by a motion controller.
Finding a collision-free, feasible motion trajectory in a complex, often high-dimensional configuration space, which is a prerequisite for autonomous robot operation.
