Loading...
Resource- and physical-constraint-aware scheduling and motion planning, for cyber-physical systems with heterogeneous processing units
McGowen, Justin
McGowen, Justin
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2023
Date Submitted
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Cyber Physical Systems (CPS) such as robots or self driving cars have strict requirements on both computation and physical operation to avoid failure. Heterogeneous (multi-accelerator) systems are also becoming commonplace in many CPS applications due to their ability to accelerate computational workloads, with different accelerators being optimal for different tasks.
A common task for such a CPS is to navigate a space, while running heavy computational workloads. We investigate how CPS performance can be improved both by considering physical constraints for workload scheduling decisions, and by considering computational decisions for motion planning.
The computation and physical operation of a CPS are intertwined---for example, physical obstacles can change computational latency requirements for braking safely, or the computational power draw might limit available power to motors. The motion of a CPS can also cause these requirements to vary.
These influences go both ways, and so performance is improved by simultaneously considering computation and physical operation. This presents two challenges. First, how do we find efficient schedules for CPS with heterogeneous processing units, such that the schedules are resource-bounded to meet the physical requirements? Second, how do we determine time or energy efficient motion paths when different paths vary not just in length, but also in the schedules satisfying the physical requirements?
We present a starting point towards addressing this problem: the Constrained Autonomous Workload Scheduler (CAuWS). CAuWS, with sufficient information and profiling prior to planning, can represent many hardware devices, environments, and constraints to generate schedules and support future motion planning work. CAuWS novelly considers physical constraints, computational constraints, and heterogeneous scheduling at the same time. By formalizing this problem, we also enable future motion planning work that considers scheduling during motion planning.
Associated Publications
Rights
Copyright of the original work is retained by the author.