CPU scheduling in robotics & AR/VR applications

Robot and AR/VR systems have to take highly responsive real-time actions, driven by complex decisions involving a pipeline of sensing, perception, planning, and reaction tasks. Given constrained resources, this leads to a difficult scheduling problem. In practice – system designers manually tune params for their specific hardware and application, while real-time scheduling approaches assume static periodic schedules – both of which result in suboptimal application performance especially when both the environment and the hardware can change. In this work, we highlight the emerging need for automated resource optimization at runtime in sense-react systems. As a step towards this goal, we identify various unique challenges in this area, especially understanding the key scheduling requirements for such systems. We propose a preliminary framework and a novel scheduling policy that enables efficient and dynamic optimization of application-specific performance goals. In experiments with a prototype implemented in the ROS and ILLIXR platforms, we show that our approach improves application performance, for example, 15x better performance for a face tracking robot and 7x better collision avoidance for a navigation robot. We believe this work will lead to systems that are substantially easier to develop and fulfill their tasks measurably better.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2021

Author dc:creator

• Aditi, -

Contributors dc:contributor

• Godfrey, Philip Brighten
• Mittal, Radhika

• Robotics
• Scheduling
• Resource management
• AR/VR

Statement dc:rights

• Copyright 2021 - Aditi

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/110582

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/110582