Dynamical Motor Control Learned with Deep Deterministic Policy Gradient

Figure 6

Curled trajectories under force field. (a) The reaching trajectories adapted for anticlockwise force field. (b) The reaching trajectories adapted for clockwise force field. Trajectories in both (a) and (b) curved to the opposite direction of the perturbing force. The short lines attached to each circle visualize the direction and the strength of the deflection force. (c) The difference between torques without force field and under force field. The torques for the two joints are arranged in rows. The torque difference demonstrates the overcompensation at the early phase of the movements.

(a)

(b)

(c)