Learning eye-arm coordination using neural and fuzzy neural techniques

Developing models for motor control, learning, and sensory-motor coordinati on is of equal interest to researchers studying human motor behavior and to engineers aiming at robots with capabilities closer to human performance, such as anthropomorphic or humanoid robots. This chapter proposes an approa ch to the transfer of motor skills to such robots, in which the robot's cap ability to control its limbs starts with the learning of motor coordination using self-directed exploration. Once it has control over its limbs, the r obot could imitate the movements of an instructor or execute movements desc ribed verbally or in a succession of coordinates in Cartesian or joint spac es. The approach described in this chapter uses learning by imitation (or t eaching by demonstration, as seen from the human perspective). This method of learning is considered promising, because it is human-friendly and effic ient in illustrating postures hard to capture in linguistic descriptions or quantified in programming instructions. The model of eye-arm sensory-motor coordination described here is characterized by a system of equations, num erically solved using neural networks. Two models of neurons are explored: a classic neuron with a sigmoidal characteristic and a fuzzy (logic) neuron defined using triangular norms, used when the model of sensory-motor coord ination was characterized by fuzzy relational equations. The interest in fu zzy neural models originates in the search for a unique structural element for a nervous system able to learn from both visual examples and linguistic descriptions of the movement. Both neural models successfully learned the functional mapping that associates images of the instructor arm with joint- commands necessary for positioning the robot arm in a similar posture. Howe ver, the fuzzy neural model has shown increased transparency. The experimen ts illustrate how the neural-guided robot is able to imitate 2D and 3D arm movements of a human and of another robot.