Date: Friday, October 17th, 2025

Time: 14:30 - 16:30 EST

Location: (virtual) https://gatech.zoom.us/j/93719106191?pwd=Fwx2p2JZumdYHFUuQBOHbkd3AB2snv.1&from=addon [Meeting ID: 937 1910 6191, Passcode: 990461]

Qikai Huang

Robotics Ph.D. Candidate

Woodruff School of Mechanical Engineering
Georgia Institute of Technology

Committee:

Dr. Seth Hutchinson (Advisor) - Khoury College of Computer Sciences, Northeastern University

Dr. Cédric Pradalier (Advisor) - School of Interactive Computing, Georgia Institute of Technology

Dr. Frank Dellaert - School of Interactive Computing, Georgia Institute of Technology

Dr. Ye Zhao - Woodruff School of Mechanical Engineering, Georgia Institute of Technology

Dr. Shreyas Kousik - Woodruff School of Mechanical Engineering, Georgia Institute of Technology

Abstract: 

The control of complex, high-degree-of-freedom robots in unstructured environments remains a significant challenge in robotics. While modern optimization-based frameworks have become a powerful tool for generating dynamic motions, their real-time performance is often hindered by a fundamental bottleneck: the computational cost of the underlying numerical solvers. General-purpose, off-the-shelf solvers are often blind to a robot's unique kinematic structure, and their computational complexity typically scales cubically with the robot's degrees of freedom and the number of constraints. This scaling imposes a practical limit on the ability to generate complex, real-time motions on high-degree-of-freedom robots in the presence of hard constraints.

This thesis directly confronts this computational challenge by developing algorithms that exploit the inherent sparsity and structure of a robot's kinematics. To support this claim, this dissertation contributes: (i) extensions to the classical operational space control paradigm that improve robustness against model uncertainties and disturbances for a class of underactuated systems; (ii) a complete planning and control pipeline for dynamic quadrupedal locomotion, including a novel convex model predictive controller that enables the execution of highly dynamic maneuvers on hardware; and (iii) a novel constrained differential inverse kinematics solver, LoIK, that employs graph elimination algorithms to achieve low computational complexity, even in the presence of complex constraints such as closed kinematic loops. Comprehensive benchmarks demonstrate that LoIK is substantially faster than state-of-the-art general-purpose solvers, helping to make real-time, optimization-based motion control a practical reality.