Reinventing Multiterminal Coding for Intelligent Machines

Introduction

Advancements in sensors and deep learning have elevated the perception capacity of machines, bringing mid-level autonomy within reach. However, the abundance of high-dimensional data, including video and dynamic point cloud streams, strains current storage and communication technologies to their limits and curtails the ability of machines to collaboratively perceive the environment, a critical factor for achieving safety and the ambitious goal of high-level autonomy.

Challenges in Current Methods

State-of-the-art cooperative perception methods are based purely on a data-driven approach, requiring massive training data and computational resources. They also lack interpretability, explainability, and a solid theoretical foundation.

Proposed Solution

This proposal puts forth a groundbreaking multiterminal coding paradigm for intelligent machines enabling data compression and communication systems that break the current limits of the predictive coding archetype. It builds a unique concept that unifies traditional distributed source coding and signal processing domain knowledge with modern deep learning.

Key Components

1. Leveraging Machine Learning:

   • It solves long-standing problems in multiterminal coding theory and devises code constructions achieving the fundamental limits.
   • Establishes a theoretical framework that defines the amount of information required to be sent per agent to solve the cooperative perception task.

2. Driving Design with Domain Knowledge:

   • It drives the design of interpretable and data- and parameter-efficient machine learning models for cooperative perception.

3. Pioneering Explanations:

   • It reinforces the interplay by pioneering explanations that enforce and assess the interpretability of the designed models.

Impact

IONIAN will have a profound impact on the way intelligent machines, including ground and aerial vehicles, and mobile robots, compress and communicate multi-sensory data to collaboratively perceive the environment for autonomous safe navigation. Ultimately, this will lead to trustworthy operation and acceptance of such systems.