Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Effective Communication with Dynamic Feature Compression

Pietro Talli, Francesco Pase, Federico Chiariotti, Andrea Zanella, Michele Zorzi

TL;DR

This work tackles remote robotic control over constrained wireless links by explicitly optimizing for semantic and effective communication. It introduces a dynamic feature compression framework based on a Deep VQ-VAE ensemble and deep reinforcement learning to adapt quantization levels in response to the environment and the receiver’s memory, demonstrating strong gains on CartPole, particularly under Level C effective control. The approach yields near-optimal control with substantially reduced bitrate and provides interpretable insights into when and why the observer should transmit, guiding design of future wireless-control systems. Compared with static or generic compression, the proposed method offers a principled, task-aware strategy for remote control in 5G-and-beyond contexts.

Abstract

The remote wireless control of industrial systems is one of the major use cases for 5G and beyond systems: in these cases, the massive amounts of sensory information that need to be shared over the wireless medium may overload even high-capacity connections. Consequently, solving the effective communication problem by optimizing the transmission strategy to discard irrelevant information can provide a significant advantage, but is often a very complex task. In this work, we consider a prototypal system in which an observer must communicate its sensory data to a robot controlling a task (e.g., a mobile robot in a factory). We then model it as a remote Partially Observable Markov Decision Process (POMDP), considering the effect of adopting semantic and effective communication-oriented solutions on the overall system performance. We split the communication problem by considering an ensemble Vector Quantized Variational Autoencoder (VQ-VAE) encoding, and train a Deep Reinforcement Learning (DRL) agent to dynamically adapt the quantization level, considering both the current state of the environment and the memory of past messages. We tested the proposed approach on the well-known CartPole reference control problem, obtaining a significant performance increase over traditional approaches.

Effective Communication with Dynamic Feature Compression

TL;DR

This work tackles remote robotic control over constrained wireless links by explicitly optimizing for semantic and effective communication. It introduces a dynamic feature compression framework based on a Deep VQ-VAE ensemble and deep reinforcement learning to adapt quantization levels in response to the environment and the receiver’s memory, demonstrating strong gains on CartPole, particularly under Level C effective control. The approach yields near-optimal control with substantially reduced bitrate and provides interpretable insights into when and why the observer should transmit, guiding design of future wireless-control systems. Compared with static or generic compression, the proposed method offers a principled, task-aware strategy for remote control in 5G-and-beyond contexts.

Abstract

The remote wireless control of industrial systems is one of the major use cases for 5G and beyond systems: in these cases, the massive amounts of sensory information that need to be shared over the wireless medium may overload even high-capacity connections. Consequently, solving the effective communication problem by optimizing the transmission strategy to discard irrelevant information can provide a significant advantage, but is often a very complex task. In this work, we consider a prototypal system in which an observer must communicate its sensory data to a robot controlling a task (e.g., a mobile robot in a factory). We then model it as a remote Partially Observable Markov Decision Process (POMDP), considering the effect of adopting semantic and effective communication-oriented solutions on the overall system performance. We split the communication problem by considering an ensemble Vector Quantized Variational Autoencoder (VQ-VAE) encoding, and train a Deep Reinforcement Learning (DRL) agent to dynamically adapt the quantization level, considering both the current state of the environment and the memory of past messages. We tested the proposed approach on the well-known CartPole reference control problem, obtaining a significant performance increase over traditional approaches.
Paper Structure (19 sections, 13 equations, 11 figures, 4 tables)

This paper contains 19 sections, 13 equations, 11 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Model
  4. POMDP Definition and Solution
  5. The Remote POMDP
  6. The Robot-Side POMDP
  7. The Observer-Side POMDP
  8. Observer Reward in Remote POMDPs
  9. Proposed Solution
  10. Deep VQ-VAE Encoding
  11. Dynamic Feature Compression
  12. RL implementation
  13. Simulation Settings and Results
  14. The Coding and Decoding Functions
  15. Neural Network Architecture and Training
  16. ...and 4 more sections

Figures (11)

  • Figure 1: Dynamic feature compression architecture.
  • Figure 2: Example of the original and reconstructed observation.
  • Figure 3: Training of the model $\zeta_6$, with 6 bits per feature.
  • Figure 4: Performance of the communication schemes on the three Levels of the remote .
  • Figure 5: Other performance metrics relative to the CartPole control problem.
  • ...and 6 more figures