The goal of this project is to computationally synthesize motion primitives for humanoids (with complex dynamics) and digital actors (using musculo-skeletal models).

We have been able to autonomously synthesize jumping motion using Probabilistic Roadmaps (PRMs).

We have essentially shown that PRMs are powerful enough to synthesize motions as complex as jumping.

To be able to jump from the configuration shown in Fig. 1(a), the leg musculo-skeletal model has to first bend at the knees as shown in Fig. 1(b). Our motion planner was able to learn this by itself as seen in the "crude" animation of the result below:

Note that the motion is not smooth in the initial phase. This is because we didnt optimized the output from the PRM planner.