RSS 2017

This week I attended the Robotics: Science and Systems (RSS) conference at MIT. I presented some work on robust trajectory optimization, where we explicitly take into account disturbances and modeling errors at the trajectory planning stage to make sure our systems can handle these things in the real world. What sets our new algorithm, DIRTREL (short for “Direct Trajectory optimization with Ellipsoidal disturbances and LQR feedback”), appart from existing methods is that it’s fast and scales up to complex real-life systems. For example, we can plan motions for things like quadrotors with unknown wind gusts and robot arms carrying containers with difficult-to-model fluid slosh. A short video of my talk from the conference is below. You can also download the full paper and check out the code on GitHub.