Agile Jumping with a 12 dof Quadruped

In this project we explored applying trajectory optimization to enabled a 12-dof quadruped to jump. By using trajectory optimization we would understand how different designs and gait sequences would effect a robot’s ability to do agile maneuvers. This work was conducted with Diego Caporale.

This project also served as an opportunity of Kod*Lab to evaluate trajectory optimization as a tool since we traditionally avoid optimization based techniques.

To start with we used Drake and DIRCON to design a pipeline for creating new jumping behaviors. The first behavior we tested was jumping mostly straight up.

Before moving to jumping a longer distance forward and also optimizing for minimum energy.

Before finally exploring different mode sequences with a bounding jump

While trajectory optimization has enabled very impressive results in the literatures, the resulting nonlinear programs are very difficult to get working well, leading to the work being part science, and part art.