Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Orchestrating Joint Offloading and Scheduling for Low-Latency Edge SLAM

Yao Zhang, Yuyi Mao, Hui Wang, Zhiwen Yu, Song Guo, Jun Zhang, Liang Wang, Bin Guo

TL;DR

This work tackles real-time vSLAM on resource-constrained mobile robots by offloading to edge servers and coordinating scheduling across multiple edge nodes. It introduces a joint architecture with regional feature prediction, an encoder/decoder configuration adaptation based on A3C DRL, and an input-dependent scheduling framework supported by a Gaussian Process to satisfy implicit, time-varying user requirements. Key contributions include tile-level importance prediction for efficient data compression, inter-server cooperation for elasticity, a QoE-centric DRL training objective, and a multi-server simulator to evaluate scheduling under realistic network dynamics. Empirical results show up to a 47% reduction in end-to-end latency and improved pose estimation accuracy while maintaining constraint satisfaction across diverse network conditions, demonstrating practical impact for scalable edge-SLAM deployments.

Abstract

Visual Simultaneous Localization and Mapping (vSLAM) is a prevailing technology for many emerging robotic applications. Achieving real-time SLAM on mobile robotic systems with limited computational resources is challenging because the complexity of SLAM algorithms increases over time. This restriction can be lifted by offloading computations to edge servers, forming the emerging paradigm of edge-assisted SLAM. Nevertheless, the exogenous and stochastic input processes affect the dynamics of the edge-assisted SLAM system. Moreover, the requirements of clients on SLAM metrics change over time, exerting implicit and time-varying effects on the system. In this paper, we aim to push the limit beyond existing edge-assist SLAM by proposing a new architecture that can handle the input-driven processes and also satisfy clients' implicit and time-varying requirements. The key innovations of our work involve a regional feature prediction method for importance-aware local data processing, a configuration adaptation policy that integrates data compression/decompression and task offloading, and an input-dependent learning framework for task scheduling with constraint satisfaction. Extensive experiments prove that our architecture improves pose estimation accuracy and saves up to 47% of communication costs compared with a popular edge-assisted SLAM system, as well as effectively satisfies the clients' requirements.

Orchestrating Joint Offloading and Scheduling for Low-Latency Edge SLAM

TL;DR

This work tackles real-time vSLAM on resource-constrained mobile robots by offloading to edge servers and coordinating scheduling across multiple edge nodes. It introduces a joint architecture with regional feature prediction, an encoder/decoder configuration adaptation based on A3C DRL, and an input-dependent scheduling framework supported by a Gaussian Process to satisfy implicit, time-varying user requirements. Key contributions include tile-level importance prediction for efficient data compression, inter-server cooperation for elasticity, a QoE-centric DRL training objective, and a multi-server simulator to evaluate scheduling under realistic network dynamics. Empirical results show up to a 47% reduction in end-to-end latency and improved pose estimation accuracy while maintaining constraint satisfaction across diverse network conditions, demonstrating practical impact for scalable edge-SLAM deployments.

Abstract

Visual Simultaneous Localization and Mapping (vSLAM) is a prevailing technology for many emerging robotic applications. Achieving real-time SLAM on mobile robotic systems with limited computational resources is challenging because the complexity of SLAM algorithms increases over time. This restriction can be lifted by offloading computations to edge servers, forming the emerging paradigm of edge-assisted SLAM. Nevertheless, the exogenous and stochastic input processes affect the dynamics of the edge-assisted SLAM system. Moreover, the requirements of clients on SLAM metrics change over time, exerting implicit and time-varying effects on the system. In this paper, we aim to push the limit beyond existing edge-assist SLAM by proposing a new architecture that can handle the input-driven processes and also satisfy clients' implicit and time-varying requirements. The key innovations of our work involve a regional feature prediction method for importance-aware local data processing, a configuration adaptation policy that integrates data compression/decompression and task offloading, and an input-dependent learning framework for task scheduling with constraint satisfaction. Extensive experiments prove that our architecture improves pose estimation accuracy and saves up to 47% of communication costs compared with a popular edge-assisted SLAM system, as well as effectively satisfies the clients' requirements.

Paper Structure

This paper contains 31 sections, 1 theorem, 15 equations, 18 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Architecture
  3. ORB-SLAM2
  4. System Architecture Overview
  5. Regional Feature Prediction Module
  6. Encoder/Decoder
  7. Configuration Adaptation Module
  8. Scheduler
  9. Algorithm Design
  10. Regional Feature Prediction
  11. Workflow of Regional Feature Prediction
  12. Training Procedure
  13. Input-Dependent Learning for Edge-Assisted SLAM
  14. MDP Formulation
  15. Configuration Adaptation Based on A3C
  16. ...and 16 more sections

Key Result

Theorem 1

Considering the optimal policy of $M_{sche.}$ and $M^-_{sche.}$ is $\pi_{sche.}$ and $\pi^-_{sche.}$, respectively, if states under $\pi^-_{sche.}$ never fall into the unsafe state region ${S^-}$, $\pi^-_{sche.}$ is the optimal policy of $M_{sche.}$, i.e., $\pi^*_{sche.}=\pi^-_{sche.}$.

Figures (18)

  • Figure 1: Proposed edge-assisted SLAM architecture.
  • Figure 2: A comparison of the extracted feature points and the generated tiles in a frame.
  • Figure 3: Workflow of the regional feature prediction module.
  • Figure 4: End-to-End Latency With 3G/HSDPA Traces
  • Figure 5: End-to-End Latency With Simulated Network Traces
  • ...and 13 more figures

Theorems & Definitions (2)

  • Definition 1
  • Theorem 1