Neuromechanics Control - Experimental data

This work combines brain-inspired and biomechanics solutions to present a neuromechanics approach for adjustable robot motor behavior.

We integrate a cerebellar spiking neural network (SNN) and a computational muscle model to enable adjustable robot stiffness and variable behavior in response to physical interactions. The cerebellar SNN provides motor adaptation and makes the controller independent of any prior modelling of the robot dynamics. The muscle model replicates mechanical viscoelasticity and includes a spinal cord reflex and variable cocontraction. By adjusting the cocontraction level, the robot stiffness can be regulated resulting in different motor behaviors when reacting to perturbations (soft vs robust behavior) and adjusting the performance accuracy. Thus, adjustable motor behavior in terms of compliance and accuracy is enabled.The neuromechanics controller operates within a feedback control loop with the Baxter robot as the front-end body.

Baxter robot is operated in torque control mode.