Memory in the Sky: Low-Altitude Question Answering with Multi-Agent Memory Aggregation
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.
Related
- Calibrated Bayesian Selection of Schema Subgraphs for Database Question Answering
- Semantic Prior Smoothing for Conformal Prediction in Multiple-Choice Question Answering with API-Only Large Language Models
- SCL: Semantic Coherence Learning for Video Question Answering
- Weighted Risk Aggregation under Long-Memory Dependence: Pathwise Fluctuations of Exposure-Weighted Losses
- Low-Altitude Fluid Antenna Network with Multi-Agent Reinforcement Learning
BibTeX
@misc{li2026memory,
title = {{Memory in the Sky: Low-Altitude Question Answering with Multi-Agent Memory Aggregation}},
author = {Chengyang Li and Yujie Wan and Shuai Wang and Kejiang Ye and Weijie Yuan and Boyu Zhou and Yik-Chung Wu and Chengzhong Xu and Huseyin Arslan},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.35431},
eprint = {2609.35431},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.35431},
}
RIS
TY - UNPB
TI - Memory in the Sky: Low-Altitude Question Answering with Multi-Agent Memory Aggregation
AU - Chengyang Li
AU - Yujie Wan
AU - Shuai Wang
AU - Kejiang Ye
AU - Weijie Yuan
AU - Boyu Zhou
AU - Yik-Chung Wu
AU - Chengzhong Xu
AU - Huseyin Arslan
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.35431
UR - https://arxiv.org/abs/2609.35431
AB - 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.
ER -