Abstract:
Robotic simulation, planning, estimation, and control, have all been built on top of numerical optimization. In this same time, modern convex optimization has matured into a robust technology delivering globally optimal solutions in polynomial time. With advances in differentiable optimization and custom solvers capable of producing smooth derivatives, convex modeling has become fast, reliable, and fully differentiable. This thesis demonstrates the effectiveness of convex modeling in areas such as Martian atmospheric entry guidance, nanosatellite space telescope pointing, collision detection, contact dynamics of point clouds, and online model learning.

Looking forward, we propose a hybrid trajectory optimization algorithm for reasoning about contact-rich manipulation tasks where derivative-free sampling is used for contact sequence discovery, and model-based optimization is used for trajectory smoothing. Together, these methods can synthesize complex manipulation behaviors in seconds without offline training required.

Thesis Committee Members:
Zachary Manchester, Chair
Zico Kolter
Changliu Liu
Tom Erez, Google DeepMind Robotics

More Information