Inverse models and robust parametric-step neuro-control of a Humanoid Robot

In this paper we present advances on the work done on the development of the AH1N2 humanoid robot at the Automatic Control Department of the Cinvestav. The geometric model of the AH1N2 humanoid robot is defined by kinematic open chains, head, arms, waist and legs, attached to the robot body. We center our work in the inverse geometric model, since from all the models used in robotics, is the most difficult to automate. Its knowledge is needed to control the robot position and attitude in the workspace. We present derivations of the inverse geometric models for the arm and the legs. We also study, in this work, the singularities of the arms and legs because they affect its control in the workspace. We also present the use of kinematic model to built movement constraints that allow us the control of the motion control and specify complex movements. Finally, we use the dynamic models to calculate a Neuro proportional-derivative control and to simulate the robot movement in presence of a destabilising perturbations. The neural network is trained to compensate the effect of gravity.