Memory in the Sky: Low-Altitude Question Answering with Multi-Agent Memory Aggregation

2026-09-28

AI Summary

This study addresses the challenges of quantifying UAV memory value and allocating collaborative resources among multiple agents in low-altitude question answering. To this end, it proposes MemCen, a novel framework that introduces an explicit memory valuation mechanism, leveraging generative adversarial examinations to assess memory quality. Furthermore, the authors derive a Quality-of-Memory (QoM)-aware water-filling algorithm to jointly optimize UAV selection and power allocation, enabling end-to-end optimization within black-box pipelines. Experimental evaluations demonstrate that the proposed approach achieves accuracies of 92.4% and 88.5% in simulated and real-world settings, respectively. These results validate the effectiveness of MemCen in facilitating environment understanding and navigation tasks for low-altitude multi-agent systems.

Abstract

This paper studies low-altitude question answering (LAQA), in which distributed unmanned aerial vehicle (UAV) memories are aggregated at a ground server to answer questions about observations over a long horizon. Unlike conventional resource allocation based on sensing, communication, control, or computation metrics, LAQA requires an explicit measure of memory value. We propose a generative adversarial exam (GAE) that uses forward simulation to evaluate memory retrieval and exam scores to quantify memory quality. This enables the downstream QA value of candidate memories to be measured and optimized without accessing the internal mechanisms of the black-box captioning, retrieval, and reasoning pipeline. Building on this metric, we develop a memory-centric (MemCen) framework that jointly selects UAVs and allocates transmit power to maximize memory quality under communication constraints. In the noise-limited regime, we derive a QoM-aware capped water-filling law that explicitly connects task utility with physical-layer power allocation. We further develop penalty successive optimization (PSO) and learning to memorize (L2M) solvers. MemCen achieves QA accuracies of 92.4% and 84.0% in CARLA Town04 and Town05 under static and dynamic communication conditions, respectively. In real-world experiments, MemCen achieves 88.5% QA accuracy on the panoramic multi-agent system (PMAS) benchmark. Finally, UAV-to-robot-dog demonstrations further validate the practical utility of the acquired memories for environmental understanding and navigation.

Problem

• Low-Altitude Question Answering
• Memory Aggregation
• Unmanned Aerial Vehicles
• Memory Value
• Resource Allocation

Innovation

• Low-Altitude Question Answering
• Generative Adversarial Exam
• Memory-Centric Framework
• Quality of Memory
• Capped Water-Filling