Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning to Optimize Edge Robotics: A Fast Integrated Perception-Motion-Communication Approach

Dan Guo, Xibin Jin, Shuai Wang, Zhigang Wen, Miaowen Wen, Chengzhong Xu

TL;DR

The paper addresses the data-upload bottleneck in edge robotics by introducing Integrated Perception, Motion, and Communication (IPMC), which jointly optimizes down-sampling, compression, and transmit power using perception- and motion-aware models. A perception-aware surrogate objective and motion-dynamics constraints are used to formulate a nonconvex problem that is recast into a geometric program (GP), solvable optimally but computationally intensive. To enable real-time operation, the authors develop Learning-to-Optimize (LTO), training a deep network offline on GP-derived demonstrations to emulate GP decisions with much lower latency. Simulation results show GP-IPMC outperforms baselines in data fidelity and efficiency, while LTO-IPMC achieves about 10× speedup with near GP performance, demonstrating practical viability for real-time, multi-modal edge robotics. The work advances ER by coupling perception, motion, and communication into a unified optimization framework with scalable, real-time solutions.

Abstract

Edge robotics involves frequent exchanges of large-volume multi-modal data. Existing methods ignore the interdependency between robotic functionalities and communication conditions, leading to excessive communication overhead. This paper revolutionizes edge robotics systems through integrated perception, motion, and communication (IPMC). As such, robots can dynamically adapt their communication strategies (i.e., compression ratio, transmission frequency, transmit power) by leveraging the knowledge of robotic perception and motion dynamics, thus reducing the need for excessive sensor data uploads. Furthermore, by leveraging the learning to optimize (LTO) paradigm, an imitation learning neural network is designed and implemented, which reduces the computational complexity by over 10x compared to state-of-the art optimization solvers. Experiments demonstrate the superiority of the proposed IPMC and the real-time execution capability of LTO.

Learning to Optimize Edge Robotics: A Fast Integrated Perception-Motion-Communication Approach

TL;DR

The paper addresses the data-upload bottleneck in edge robotics by introducing Integrated Perception, Motion, and Communication (IPMC), which jointly optimizes down-sampling, compression, and transmit power using perception- and motion-aware models. A perception-aware surrogate objective and motion-dynamics constraints are used to formulate a nonconvex problem that is recast into a geometric program (GP), solvable optimally but computationally intensive. To enable real-time operation, the authors develop Learning-to-Optimize (LTO), training a deep network offline on GP-derived demonstrations to emulate GP decisions with much lower latency. Simulation results show GP-IPMC outperforms baselines in data fidelity and efficiency, while LTO-IPMC achieves about 10× speedup with near GP performance, demonstrating practical viability for real-time, multi-modal edge robotics. The work advances ER by coupling perception, motion, and communication into a unified optimization framework with scalable, real-time solutions.

Abstract

Edge robotics involves frequent exchanges of large-volume multi-modal data. Existing methods ignore the interdependency between robotic functionalities and communication conditions, leading to excessive communication overhead. This paper revolutionizes edge robotics systems through integrated perception, motion, and communication (IPMC). As such, robots can dynamically adapt their communication strategies (i.e., compression ratio, transmission frequency, transmit power) by leveraging the knowledge of robotic perception and motion dynamics, thus reducing the need for excessive sensor data uploads. Furthermore, by leveraging the learning to optimize (LTO) paradigm, an imitation learning neural network is designed and implemented, which reduces the computational complexity by over 10x compared to state-of-the art optimization solvers. Experiments demonstrate the superiority of the proposed IPMC and the real-time execution capability of LTO.
Paper Structure (11 sections, 17 equations, 5 figures)

This paper contains 11 sections, 17 equations, 5 figures.

Table of Contents

  1. Introduction
  2. System Model
  3. Integrated Perception, Motion, and Communication for Edge Robotics
  4. Perception-Aware Objective Function
  5. Motion-Dynamics Aware Communication Constraints
  6. IPMC Problem Formulation
  7. Learning-to-Optimize Algorithm
  8. Geometric Programming
  9. Learning to Optimize
  10. Simulation Results
  11. Conclusion

Figures (5)

  • Figure 1: System model of edge robotics and proposed perception-motion-communication models.
  • Figure 2: Pipeline of the LTO-IPMC algorithm.
  • Figure 3: Quantitative comparison of PSNRs and MSEs.
  • Figure 4: Visualization of reconstructed camera and lidar data.
  • Figure 5: a) Training and test accuracies of LTO; b) Training and test losses of LTO; c) MSE versus $P$ for GP-IPMC and LTO-IPMC; d) PSNR versus $P$ for GP-IPMC and LTO-IPMC; e) Comparison of computation time.