We are thrilled to congratulate Dr. Anuj Pasricha, who successfully defended his PhD dissertation titled “Expanding Robot Manipulation Capability Through Dynamics-Constrained Motion Generation” on May 19, 2026. Anuj was advised by Alessandro Roncone. Congratulations Dr. Pasricha!
His dissertation PDF is available [here].
Abstract
Robots operating in the physical world must generate motions that are not only directed toward a task objective, but also admissible under the kinematic and dynamic constraints of the robot-world system. In manipulation, these constraints—for instance, contact geometry, friction, actuation limits, and object inertia—determine not only which states are reachable, but also which transitions are physically realizable and which regions of the motion space are meaningful to search. This thesis develops traditional and learning-based planning methods for manipulation problems in which both kinematic and dynamic constraints shape the set of achievable behaviors.
In developing these methods, my work expands the scope of robotic manipulation beyond geometric pick-and-place by incorporating dynamic non-prehensile skills and motions governed by object inertial properties, impulsive contact, and motion-induced forces. First, it introduces dynamics-aware planning methods that evaluate admissibility online through kinodynamic primitives, physics simulation, and control-aware search, enabling manipulation behaviors involving impulsive contact, coupled robot-object dynamics, and liquid transport. Second, it develops reusable kinodynamic planning representations that amortize repeated feasibility computation while preserving explicit dynamically feasible motion structure across related manipulation queries. Third, it transfers admissibility knowledge into conditional generative models that rapidly sample trajectories conditioned on task, embodiment, and physical parameters such as payloads and actuation limits. Fourth, it reframes constraint identification in terms of induced behavior, using anchor probes to compare constraints by behavioral equivalence rather than by perceptual or parameter similarity alone. Together, these contributions expand the range of manipulation behaviors that robotic systems can plan for and execute, and provide methods for doing so efficiently as task conditions, embodiments, and physical parameters of the robot-world system change.
Bio
Anuj Pasricha is a Senior Software Engineer, Robotics R&D at Berkshire Grey. His research expands the scope of robotic manipulation beyond geometric reasoning by developing methods that model and plan over dynamic and non-prehensile interactions, enabling new capabilities governed by object inertial properties, impulsive contact, and motion-induced forces.