mihail  pivtoraiko

We investigated application of covariant optimization planning to manipulation, including grasping, as well as pick and place operations.

We proposed an approach to efficient manipulator base placement that simultaneously satisfies arbitrary mobility constraints of the base, admits efficient search methods, and maximizes a user-defined measure of manipulation quality.

My thesis research focuses on developing search spaces for efficient kinodynamic motion planning. The proposed state lattice search spaces feature motion primitives that are compatible with some of the most efficient search techniques, while guaranteeing  satisfaction of differential constraints.

Autonomous goal acquisition is an important capability in many applications, including planetary exploration. By developing efficient representations of robot mobility, we proposed approaches that outperform the state-of-the-art methods in planetary rover navigation.

I lead one of the two directions of research on this project, namely the development of optimal regional motion planning under mobility constraints. The project focused on robots with stringent power and computational limitations, such as  planetary exploration rovers. The developed planner judiciously exploited rover mobility, including point turns and sideways crabbing motions, while offering the capacity to minimize path length, energy expenditure and other relevant measures of traversing rough terrain.

On a tight schedule, I ported the PerceptOR navigation and planning system to the LAGR robots, while they were being shaken out for delivery to the LAGR kickoff event. This included the adaptation of the system from car-like to differential drive mobility. The ported code became part of the standard LAGR baseline system.

As part of my involvement, I implemented the Plan-Around behavior of the robot that was invoked to get the robot out of toughest spots it encountered while roaming forests, ravines and other difficult terrain. This was PerceptOR's last resort to keep moving.

Reversible logic synthesis is motivated by its capacity to provide theoretical support for constructing circuits without the loss of information, and consequently, without energy loss. We proposed an automated synthesis of reversible logic circuits in the form of cascades of generalized gates using genetic algorithms.