Communication-Aware Iterative Map Compression for Online Path-Planning

TL;DR

The paper tackles online path planning for multi-robot teams under limited communication by introducing a communication-aware map compression framework. A Kalman-filter–based iterative Decoder estimates the unknown map from a set of abstractions transmitted by a mobile Sensor, while an Encoder selects the optimal abstraction $ heta_t$ to minimize a joint communication- and estimation-based objective. The Path Planner uses mean estimates to compute routes, and a Path Converter focuses processing around the planned path to guide abstraction selection. Across Mars and Earth experiments, the approach achieves up to a $98\%$ reduction in transmitted data with comparable path costs and reduced Decoder time compared to prior methods, enabling larger abstraction templates and more challenging scenarios.

Abstract

This paper addresses the problem of optimizing communicated information among heterogeneous, resource-aware robot teams to facilitate their navigation. In such operations, a mobile robot compresses its local map to assist another robot in reaching a target within an uncharted environment. The primary challenge lies in ensuring that the map compression step balances network load while transmitting only the most essential information for effective navigation. We propose a communication framework that sequentially selects the optimal map compression in a task-driven, communication-aware manner. It introduces a decoder capable of iterative map estimation, handling noise through Kalman filter techniques. The computational speed of our decoder allows for a larger compression template set compared to previous methods, and enables applications in more challenging environments. Specifically, our simulations demonstrate a remarkable 98% reduction in communicated information, compared to a framework that transmits the raw data, on a large Mars inclination map and an Earth map, all while maintaining similar planning costs. Furthermore, our method significantly reduces computational time compared to the state-of-the-art approach.

Figures (6)

• Figure 1: Earth traversibility map $\mathcal{M}$. The mobile robot perceives a local map of dimensions $w \times h$ as it traverses the map and compresses it based on an abstraction template.
• Figure 2: Example of a set of abstractions. The grey area denotes no information.
• Figure 3: Proposed framework schematic. The Actor shares its current optimal path $\pi^*_t$ to the Sensor. The Sensor uses this information to select the optimal abstraction $\theta^*_t \in \Theta$ of its local map and sends it to the Actor.
• Figure 4: (a) Discretized environment with static obstacles. $\textbf{S}$ denotes the starting cell while $\textbf{G}$ denotes the target; (b) Associated graph of the environment employing the set of actions $U =$ {UP, DOWN, LEFT, RIGHT}, along with the optimal path (blue nodes) with cost function \ref{['eq:cell_cost']}; (c) Path weights computed using \ref{['eq:path_weights']}.
• Figure 5: (a) Mars inclination map computed from the depth map in HiRISE2024; (b-d) Final estimated map with $v^A \sim \mathcal{N}(0,10^{-5}I)$ for (b) Uninformed Framework, (c) Abstraction-Selection Framework, (d) Fully-Informed Framework. Blue and red indicate the cells that the Actor and the Sensor have already traversed, respectively.