Robotic Arm Bayesian Programming

PhD thesis of Rubén Senen Garcia Ramirez (2003)

This thesis focuses on Bayesian programming of manipulator arms equipped with a stereoscopic vision system.

The implementation of an object pick-up and drop-off task is given as an example to evaluate the approach. This task involves geometric models of the arm, the stereoscopic vision system and the manipulated objects. Uncertainty associated with the geometric models, robot sensors, and vision system is considered.

The programming method is formal and systematic. It consists of three parts:

• The description where the relevant variables and the joint probability distribution are defined.
• The inference which aims at building a probability distribution on the motor commands knowing the sensory information,
• and finally the choice of a command from the previous distribution.

A new method of robot programming called “inverse” is introduced. Approximate resolution methods are presented aiming to combat the computational complexity associated with this approach. A qualitative experiment of the programming system is presented.