Figure 8.

This figure illustrates the controls of best learned policies for the 4-link balancing task using DMPs (A) and PMPs (B). The corresponding joint angle trajectories are shown in Figure 7. The controls were perturbed by zero-mean Gaussian noise with σ = 10 Nm. Shown are 100 roll-outs using the same parameter setting θ for each approach. The dotted horizontal lines in the last row indicate the control constraints.