Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Maze Discovery using Multiple Robots via Federated Learning

Kalpana Ranasinghe, H. P. Madushanka, Rafaela Scaciota, Sumudu Samarakoon, Mehdi Bennis

TL;DR

The paper tackles the generalization gap in maze-shape classification for multi-robot navigation by applying federated learning (FedAvg) across two LiDAR-equipped robots operating in distinct square-grid mazes. It demonstrates that locally trained models fail to generalize to unseen wall geometries, while federated collaboration yields a robust global model that accurately classifies block types in both mazes (≈99% accuracy). The approach leverages edge servers for visualization and uses a consistent neural network architecture with well-defined navigation, data collection, and training protocols. The work highlights FL's practical value for robust, privacy-preserving, cross-environment perception and navigation in real-world multi-robot systems, with a concrete demonstration on LiDAR-based maze discovery and future directions toward topology-aware LiDAR features.

Abstract

This work presents a use case of federated learning (FL) applied to discovering a maze with LiDAR sensors-equipped robots. Goal here is to train classification models to accurately identify the shapes of grid areas within two different square mazes made up with irregular shaped walls. Due to the use of different shapes for the walls, a classification model trained in one maze that captures its structure does not generalize for the other. This issue is resolved by adopting FL framework between the robots that explore only one maze so that the collective knowledge allows them to operate accurately in the unseen maze. This illustrates the effectiveness of FL in real-world applications in terms of enhancing classification accuracy and robustness in maze discovery tasks.

Maze Discovery using Multiple Robots via Federated Learning

TL;DR

The paper tackles the generalization gap in maze-shape classification for multi-robot navigation by applying federated learning (FedAvg) across two LiDAR-equipped robots operating in distinct square-grid mazes. It demonstrates that locally trained models fail to generalize to unseen wall geometries, while federated collaboration yields a robust global model that accurately classifies block types in both mazes (≈99% accuracy). The approach leverages edge servers for visualization and uses a consistent neural network architecture with well-defined navigation, data collection, and training protocols. The work highlights FL's practical value for robust, privacy-preserving, cross-environment perception and navigation in real-world multi-robot systems, with a concrete demonstration on LiDAR-based maze discovery and future directions toward topology-aware LiDAR features.

Abstract

This work presents a use case of federated learning (FL) applied to discovering a maze with LiDAR sensors-equipped robots. Goal here is to train classification models to accurately identify the shapes of grid areas within two different square mazes made up with irregular shaped walls. Due to the use of different shapes for the walls, a classification model trained in one maze that captures its structure does not generalize for the other. This issue is resolved by adopting FL framework between the robots that explore only one maze so that the collective knowledge allows them to operate accurately in the unseen maze. This illustrates the effectiveness of FL in real-world applications in terms of enhancing classification accuracy and robustness in maze discovery tasks.
Paper Structure (9 sections, 4 figures)

This paper contains 9 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Features of Square Grid Mazes
  3. System Architecture & Hardware
  4. System Implementation
  5. Navigation between adjacent grids
  6. Data Collection
  7. Training the classification model
  8. Maze discovery
  9. Demo, Resources, and Future Developments

Figures (4)

  • Figure 1: Block types present in a maze with respect to the orientation of the observer and their corresponding label.
  • Figure 2: Robotic platform of two mazes.
  • Figure 3: Discovering mazes (top) using locally trained models: inference output of the own maze (middle) vs the other maze (bottom).
  • Figure 4: Discovering mazes (top) using FL models: inference output of the own maze (middle) vs the other maze (bottom).