Loading...
Thumbnail Image
Publication

Motion planning with task scheduling in heterogeneous computing systems

Fields, Noah B.
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Motion planning is an important problem in many contexts of robotics. Heterogeneous computing systems in robots are able to run tasks on different processing units in varying orders, but with different impacts on the robot's state and performance. Currently existing sampling-based motion planning frameworks explore a state space through typically random sampling to create a path to a goal region, but only consider physical obstacles in their way such as walls, and do not consider the constraints of computational requirements on the path or the impacts of choosing different schedules for computation. We introduce a novel system which uses Petri nets as a modeling system on the computational requirements, and uses constraint solvers to find computation schedules for the motion planning tasks. This allows us to select motions based not only on their physical validity, but also computation-related parameters. We subdivide a space of constraints on the system into regions, enabling schedule reuse in order to improve the algorithm's efficiency. We also discuss the use of Petri nets to model another aspect of computation in a heterogeneous environment, memory contention. Our system enables us to consider physical dynamics such as heat and power in a way that prior systems are not capable. We demonstrate that our system can handle a variety of constraints of different severities, and can avoid computational obstacles more effectively than na\"ive planning systems which do not consider computational constraints and instead disallow regions as though they are physical obstacles.
Associated Publications
Rights
Copyright of the original work is retained by the author.
Embedded videos