Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Potential gains of communication-compute-control co-design based performance optimization methods in cyber-physical systems

Sándor Rácz, Norbert Reider

TL;DR

The paper tackles the challenge of non-determinism in cyber-physical systems arising from wireless and cloud computing by introducing three co-design-based performance optimization methods: adaptive command types, AI-driven extrapolation, and trajectory time-scaling. These methods leverage knowledge of communication and compute gaps to reduce trajectory deviation during open-loop intervals and to shorten overall execution time. Quantitative evaluation on a UR5e robotic arm controlled from an edge cloud over a private 5G network shows that AI-based extrapolation can yield an order-of-magnitude improvements in trajectory extrapolation accuracy and that combining methods leads to up to 45% shorter trajectories, with additive gains observed when integrating the approaches. The results suggest that app-level co-design, with lightweight modeling of imperfections as gaps, can significantly enhance performance without exorbitant resource investments, enabling robust, cloud-native, and wireless-enabled control of robotic systems.

Abstract

In this paper we propose and quantitatively evaluate three performance optimization methods that exploit the concept of communication-compute-control co-design by introducing awareness of communication and compute characteristics into the application logic in different ways to improve overall system performance. We have implemented a closed-loop control of a robotic arm over a wireless network where the controller is deployed into an edge cloud environment. When implementing an industrial system that leverages network and cloud technologies, the level of determinism of the control application can be decreased by nature. This means that some imperfections may be introduced into the control system, and the closed-loop control in substance changes to open-loop during disturbances. We aim to improve the performance of these open-loop control periods by applying methods that can compensate for the imperfections statistically or in a guaranteed way. We demonstrate that co-design-based application improvements with minimal dependencies on the underlying technologies can already yield an order of magnitude gain when it comes to the accurate execution of the robot trajectories during the openloop control periods. Furthermore, by combining the proposed methods, the performance improvements add up and can produce up to 45% shorter trajectory executions compared to individual evaluations.

Potential gains of communication-compute-control co-design based performance optimization methods in cyber-physical systems

TL;DR

The paper tackles the challenge of non-determinism in cyber-physical systems arising from wireless and cloud computing by introducing three co-design-based performance optimization methods: adaptive command types, AI-driven extrapolation, and trajectory time-scaling. These methods leverage knowledge of communication and compute gaps to reduce trajectory deviation during open-loop intervals and to shorten overall execution time. Quantitative evaluation on a UR5e robotic arm controlled from an edge cloud over a private 5G network shows that AI-based extrapolation can yield an order-of-magnitude improvements in trajectory extrapolation accuracy and that combining methods leads to up to 45% shorter trajectories, with additive gains observed when integrating the approaches. The results suggest that app-level co-design, with lightweight modeling of imperfections as gaps, can significantly enhance performance without exorbitant resource investments, enabling robust, cloud-native, and wireless-enabled control of robotic systems.

Abstract

In this paper we propose and quantitatively evaluate three performance optimization methods that exploit the concept of communication-compute-control co-design by introducing awareness of communication and compute characteristics into the application logic in different ways to improve overall system performance. We have implemented a closed-loop control of a robotic arm over a wireless network where the controller is deployed into an edge cloud environment. When implementing an industrial system that leverages network and cloud technologies, the level of determinism of the control application can be decreased by nature. This means that some imperfections may be introduced into the control system, and the closed-loop control in substance changes to open-loop during disturbances. We aim to improve the performance of these open-loop control periods by applying methods that can compensate for the imperfections statistically or in a guaranteed way. We demonstrate that co-design-based application improvements with minimal dependencies on the underlying technologies can already yield an order of magnitude gain when it comes to the accurate execution of the robot trajectories during the openloop control periods. Furthermore, by combining the proposed methods, the performance improvements add up and can produce up to 45% shorter trajectory executions compared to individual evaluations.

Paper Structure

This paper contains 31 sections, 3 equations, 15 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. 5G technology
  3. Related work
  4. Related work from the area of NCS and WNCS
  5. Related work from cloud domain
  6. Pragmatic co-design approaches
  7. Conceptual and theoretical work on co-design
  8. Relevance of co-design
  9. Scope and contributions
  10. System model
  11. Assumptions
  12. System architecture
  13. Proposed performance optimization methods
  14. Method-A - Adaptive command type
  15. Method-B - AI-based extrapolation
  16. ...and 16 more sections

Figures (15)

  • Figure 1: Illustration of the functional system architecture combined with the 5G SA private network deployment used for the measurements
  • Figure 2: Illustration on the impact of a gap on the trajectory execution
  • Figure 3: Location of the proposed components
  • Figure 4: Illustration the robot behavior during gap for speedj and speedl commands. After an initial period, the two trajectories deviate from the planned trajectory following different paths.
  • Figure 5: Illustration of the measured deviation from the planned trajectory for only speedj, only speedl commands and the adaptive method. The control loop is intentionally not tuned to highlights and enlarge transient periods.
  • ...and 10 more figures

Theorems & Definitions (1)

  • Definition 2.1: Imperfections