β-Sparse GP framework cuts robot path cost 18%, data transmission 76%

A team led by Evangelos Psomiadis, Dipankar Maity, and Panagiotis Tsiotras has introduced a novel framework for collaborative navigation of heterogeneous robots in unknown environments. The setup involves a lead robot navigating toward a target while a mobile sensor robot (e.g., a drone) assists by transmitting locally observed environmental data under strict bandwidth constraints. The core innovation is β-Sparse Gaussian Processes, a robust variational sparse GP model designed for task-aware inducing point selection. This allows the sensor to simultaneously choose which map points to transmit and which navigation actions to take, all while predicting unexplored regions. An action-selection strategy balances task relevance with exploration, making the system adaptive to real-world constraints.

Simulations using Mars and Earth terrain maps demonstrate significant performance gains: the framework reduces path cost by 18% compared to scenarios with no inter-robot communication, and decreases transmitted information by 76% relative to raw-data transmission baselines. These results highlight the potential for deploying such intelligent, bandwidth-aware coordination in planetary rovers, search-and-rescue drones, and autonomous underground mapping. By combining probabilistic modeling with online decision-making, the approach paves the way for more efficient multi-agent exploration in communication-constrained environments.