Dynamical Motor Control Learned with Deep Deterministic Policy Gradient

Figure 5

Training and validation of the reaching movement generation. (a) Velocity profiles of reaching movements in different directions. The plot of hand velocity versus time of all the center-out movements displayed a bell-shaped profile. Color coding was as in Figure 4(c). (b) The activation of six muscles for 16 reaching directions (each panel for a muscle). Horizontal and vertical axes represent time and reaching direction, respectively. (c) Activation of a typical neuron in the dynamical controller for movements in different directions.

(a)

(b)

(c)