Zac Manchester Engineer, Rocket Scientist, Spacecraft Hacker

ISRR 2017

This week I’m attending the 2017 International Symposium on Robotics Research (ISRR) in Puerto Varas, Chile. Yesterday, I presented some work on motion planning for systems that make and break contact with their environment. This is a really pervasive problem in robotics that shows up in both walking and manipulation. Fundamentally, it’s challenging because impacts and friction lead to trajectories with discontinuities (large jumps in velocity). These discontinuities break a lot of traditional algorithms that rely on derivatives and assume some level of smoothness.

We developed a new, more accurate, mathematical formulation of rigid body contact physics that works well for trajectory optimization. Instead of explicitly reasoning about impact and friction forces, our formulation treats the physics as a constrained optimization problem, where forces are generated implicitly by inequality constraints that, for example, say that objects are not allowed to fall through the floor. Using this approach, we can generate walking trajectories for legged robots like Spring Flamingo without having to pre-specify a contact sequence. You can check out the slides from my talk, download the full paper and get the code on GitHub.