Robot motion planning by limited space method
Mladen Crneković, Davor Zorc, Dubravko Majetić · University of Zagreb University Computing Centre (SRCE) · 2002
Paper [1] starts by the fact that two objects in a space can be in 6 different positions relative to each other.Apart from the condition that the objects must be convex, there are no limitations in the number of plains the objects are composed of.Typical bodies are investigated (cubes, rectangular boxes, cylinders, cones) that can build up a robot geometry model.The proposed algorithm is fast, and after initialization it gives the Euclidean distance of the objects often in 3 to 4 ms, but it does not calculate any path.Paper [2] presents an algorithm which replaces objects with a set of spheres with different size and distribution.It can simulate even concave objects, but generates pseudo-optimal solutions for collision detection.Solution time for the collision detection is from 1.1 to 2.8 seconds (up to 10.5 s for »normal hard« model).As in reference [1], the algorithm does not calculate a robot path.To reduce the solution time, the authors have decided to use parallel algorithm and up to 11 processors.Instead of using planes, paper [3] defines possible collision points (two spheres per robot link), therefore simplifies the problem.The concept exploits the recursive forward kinematic structure and is not based on the configuration space representation.Although 10-axis manipulator is modeled, there is no information about solution times.