Georgia Institute of Technology
Structure-Aware Task and Motion Planning for Contact-rich Locomotion and Manipulation
Abstract
dc:description.abstractContact-rich locomotion and manipulation underpin a wide range of next-generation robotic applications, from quadrupedal navigation through unstructured terrain to whole-body collaborative manipulation by legged manipulators and humanoids. Realizing these capabilities at scale requires planning and control frameworks that are simultaneously efficient, agile, safe, and scalable—a goal complicated by high-dimensional and hybrid dynamics, the combinatorial structure of contact, and the persistent gap between high-level task decisions and low-level physical feasibility. This dissertation develops a unified planning-and-control framework for contact-rich locomotion and manipulation, organized around a single technical motif: every problem admits a structure-aware decomposition that is solved through alternating coordination between subproblems. The motif is instantiated in four complementary decomposition patterns. First, a temporal decomposition splits contact-rich trajectory optimization into per-timestep blocks coordinated by multi-block alternating direction method of multipliers (ADMM) consensus, with a stage-wise accelerated variant validated on a kneed-compass-gait biped and a real-time model predictive control (MPC) variant validated on a 7-DoF KUKA manipulator pressing against a compliant substrate. Second, a spatial decomposition exploits the star-shaped coupling induced by a shared payload to produce a distributed MPC for multi-robot collaborative loco-manipulation whose re-planning rate is essentially independent of team size. Third, a system-level decomposition between a momentum-inertia-aware centroidal block—parameterizing time-varying centroidal inertia through an equimomental ellipsoid—and a whole-body block, closed by a dynamics-consensus loop, generates agile, momentum-rich motions that single-rigid-body pipelines cannot produce; a full-centroidal MPC tracks the resulting trajectories on the MIT Mini Cheetah with a 180-degree twisting-jump maneuver. Fourth, a cross-layer decomposition across the task–motion interface unifies linear-temporal-logic-based (LTL-based) reactive synthesis with mixed-integer convex programming (MICP) as a physical-feasibility certificate, closing the symbolic-to-continuous-planning gap for terrain-adaptive locomotion and validated on two quadrupedal platforms over unstructured and rebar terrains. Together, these four patterns deliver efficient, agile, safe, and scalable planning and control for contact-rich locomotion and manipulation, with rigorous simulation and hardware validation spanning a wide range of robotic platforms. The unifying lens of structure-aware decomposition with alternating coordination further offers a methodological foundation for future work at the intersection of model-based optimization, task planning, and learning-based control.
Degree
thesis:*- Name thesis:degree_name
- Electrical and Computer Engineering, PhD
- Department thesis:degree_department
- Electrical and Computer Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhou, Ziyi
- Advisor dc:contributor.advisor
-
- Zhao, Ye
- Committee members dc:contributor.committeemember
-
- Hutchinson, Seth
- Coogan, Samuel
- Wensing, Patrick
- ORCID person:identifier.orcid
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/82114