A model-based approach to robot kinematics and control using discrete factor graphs with belief propagation

Much of recent researches in robotics have shifted the focus from traditionally-specific industrial tasks to investigations of new types of robots with alternative ways of controlling them. In this paper, we describe the development of a generic method based on factor graphs to model robot kinematics. We focused on the kinematics aspect of robot control because it provides a fast and systematic solution for the robot agent to move in a dynamic environment. We developed neurally-inspired factor graph models that can be applied on two different robotic systems: a mobile platform and a robotic arm. We also demonstrated that we can extend the static model of the robotic arm into a dynamic model useful for imitating natural movements of a human hand. We tested our methods in a simulation environment as well as in scenarios involving real robots. The experimental results proved the flexibility of our proposed methods in terms of remodeling and learning, which enabled the modeled robot to perform reliably during the execution of given tasks. Neurally inspired factor graphs for generic kinematic modeling of robot platforms.Performance evaluation on a real robot arm and an omnidirectional mobile robot.Demonstration of a dynamic factor graph for imitation learning.