Robotics Research Changes the Trajectory of Drones for the Better

For years, Illinois Computer Science professor Kris Hauser has investigated how multi-level optimization techniques could be applied to robot manipulation and legged robots. Until recently, he hadn’t considered whether those techniques could apply to agile unmanned aerial vehicles (UAVs), a.k.a. drones.

Along with his PhD student, Gao Tang, and Weidong (Bill) Sun, a master’s student at Duke University, Hauser showed how this technique improved the way drones navigate around obstacles compared to previous methodology. Their discoveries were published in a paper titled “Fast UAV Trajectory Optimization using Bilevel Optimization with Analytical Gradients”.

The importance of this work was recognized recently by the IEEE Technical Committee for Model-Based Optimization for Robotics, which granted Hauser’s collaborative research a Best Paper Award for 2021.

“The basic idea is that you can separately optimize the shape of the path and the speed along it, but since how fast you can fly depends on the shape of the path, these two components are intimately coupled,” Hauser said. “Our novel formulation treats this as a bilevel optimization problem, in which the speed optimization is nested inside the shape optimization as an inner subproblem. We derived a technique for efficiently determining gradients of the optimum of the inner problem so that an optimizer for the outer problem can proceed efficiently.

“Our experiments showed that this technique is faster and significantly more numerically stable than optimizing both the shape and the speed simultaneously.”

With this recognition from the IEEE Technical Committee for Model-Based Optimization for Robotics, Hauser and his team were honored by a community devoted to “fostering innovation in optimization methods for implementing dynamic behaviors in robotics.”

Hauser said the committee recognizes a journal or conference paper from multiple robotics publications each year for its Best Paper Award.

The committee states that the award is meant to “acknowledge and promote exceptional papers published during a calendar year and recognize the contribution of their authors to the TC topics.”

To the professor, this acknowledgement validates a project that he already believed in, regarding the merits of its value to the robotics community.

“The most meaningful aspect of this paper is that if you have a hard optimization problem that can be split into easier problems with a nested structure, bilevel optimization with analytical gradients lets you solve them efficiently and reliably,” Hauser said. “We happened to identify that simultaneous shape and speed optimization for UAVs satisfies this property and leads to a very practical algorithm.

“So much credit goes to Gao and Weidong for their dedicated work that led to an innovative approach. It’s a pleasure to work with such students on topics that truly make a difference in robotics research.”