Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimal Control Selection over the Edge-Cloud Continuum

Xiyu Gu, Matthias Pezzutto, Luca Schenato, Subhrakanti Dey

TL;DR

This work addresses real-time control over the edge-cloud continuum by introducing a three-tiered controller (onboard, edge, cloud) with a delay-compensation network and a predictive, cost-driven control selection policy. It formalizes the computing continuum, models non-ideal network delays, and defines a finite-horizon optimization that selects among available control sequences, including a robust onboard fallback. The authors prove Input-to-State Stability under bounded disturbances and demonstrate substantial performance gains through simulations on a mobile-robot task, showing robustness to packet losses and delays. The approach enables seamless orchestration of multiple computing resources to improve control performance while maintaining safety, with practical implications for industrial and robotic applications on heterogeneous networks.

Abstract

The emerging computing continuum paves the way for exploiting multiple computing devices, ranging from the edge to the cloud, to implement the control algorithm. Different computing units over the continuum are characterized by different computational capabilities and communication latencies, thus resulting in different control performances and advocating for an effective trade-off. To this end, in this work, we first introduce a multi-tiered controller and we propose a simple network delay compensator. Then we propose a control selection policy to optimize the control cost taking into account the delay and the disturbances. We theoretically investigate the stability of the switching system resulting from the proposed control selection policy. Accurate simulations show the improvements of the considered setup.

Optimal Control Selection over the Edge-Cloud Continuum

TL;DR

This work addresses real-time control over the edge-cloud continuum by introducing a three-tiered controller (onboard, edge, cloud) with a delay-compensation network and a predictive, cost-driven control selection policy. It formalizes the computing continuum, models non-ideal network delays, and defines a finite-horizon optimization that selects among available control sequences, including a robust onboard fallback. The authors prove Input-to-State Stability under bounded disturbances and demonstrate substantial performance gains through simulations on a mobile-robot task, showing robustness to packet losses and delays. The approach enables seamless orchestration of multiple computing resources to improve control performance while maintaining safety, with practical implications for industrial and robotic applications on heterogeneous networks.

Abstract

The emerging computing continuum paves the way for exploiting multiple computing devices, ranging from the edge to the cloud, to implement the control algorithm. Different computing units over the continuum are characterized by different computational capabilities and communication latencies, thus resulting in different control performances and advocating for an effective trade-off. To this end, in this work, we first introduce a multi-tiered controller and we propose a simple network delay compensator. Then we propose a control selection policy to optimize the control cost taking into account the delay and the disturbances. We theoretically investigate the stability of the switching system resulting from the proposed control selection policy. Accurate simulations show the improvements of the considered setup.

Paper Structure

This paper contains 12 sections, 2 theorems, 38 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Computing Continuum model
  4. System model
  5. Network model
  6. Objective
  7. Proposed strategy
  8. Network compensator
  9. Selection policy
  10. Stability analysis
  11. Simulations
  12. Conclusions

Key Result

Lemma 1

Let $d(k)$ be a random delay process. For any $\rho > 0$ there exists a $D>0$ such that $Pr(d(k) > D) < \rho$. Moreover: where $\Phi(\cdot)$ is cumulative distribution function of $\mathcal{N}(0,1)$.

Figures (3)

  • Figure 1: Computing Continuum
  • Figure 2: System trajectories (top) and cost degradation with 100 runs Monte Carlo test (bottom) varying channel conditions without disturbances.
  • Figure 3: System trajectories with disturbance.

Theorems & Definitions (3)

  • Lemma 1
  • Theorem 1
  • Proof 1