The results support TriCLE as a practical prototype for interpretable, edge-feasible aircraft grouping, while emphasizing the need for further validation on real aligned thermal and LiDAR sensor streams.
Abstract
Edge platforms used for aerial observation must interpret aircraft imagery under limited memory, limited compute, and intermittent connectivity. This setting is difficult for standard RGB-only recognition models and general-purpose vision-language models, especially when calibrated thermal and LiDAR aircraft data are unavailable. We present TriCLE, an application-oriented tri-modal vision-language system for aircraft taxonomic grouping under edge constraints. From a single RGB aircraft image, TriCLE generates a structure-preserving FLIR-style thermal view and a pseudo-LiDAR depth projection, then fuses the aligned views with task instructions in a compact Qwen3-VL backbone. The model is aligned to an expert aircraft taxonomy based on propulsion, airframe family, size, design era, and configuration, so its outputs reflect engineering-relevant similarity rather than only surface appearance. We evaluate supervised fine-tuning, rotation-preserving SFT, and three policy-alignment strategies: GRPO, GSPO, and DAPO. Sequence-level GSPO gives the strongest validation performance, reaching 88.33\% validation accuracy and 0.91 weighted F1 on valid aircraft outputs. On a held-out aircraft test partition, GSPO achieves 78.00\% accuracy and 0.793 weighted F1 while preserving 94.00\% parseable output formatting. After 4-bit quantization and attention-memory optimization, the aligned 4B model fits an 8GB deployment target and processes each tri-modal triplet in 1.48 seconds. These results support TriCLE as a practical prototype for interpretable, edge-feasible aircraft grouping, while emphasizing the need for further validation on real aligned thermal and LiDAR sensor streams.
Unmanned aerial vehicles (UAVs) are used for a wide range of tasks, many of which are autonomous. Since UAVs operate in three-dimensional space, visual data presents a rich source of input for them. Meanwhile, the rapid development of artificial intelligence has given rise to vision-language models (VLMs). These models not only recognize objects in a scene, as traditional computer vision techniques do, but also demonstrate a strong understanding of scene context, enabled by large language models (LLMs) serving as one of their core components. However, applications as specialized as aerial robotics typically require custom-trained models, which demand significant computational resources, large datasets, and expensive hardware. Motivated by these limitations, we investigate whether a state-of-the-art, general-purpose VLM can be leveraged for UAV-relevant tasks through carefully designed prompts, without additional training. We evaluate the VLM’s understanding of aerial scenes through the task of image captioning, using visual question answering that targets spatial information within the scene. Furthermore, we utilize VLMs to automatically generate metadata for aerial datasets, enabling more complex future tasks. Results show that VLMs can generate accurate metadata with high agreement with human-annotated data even without retraining. They also demonstrate that careful prompt engineering enhances the model’s ability to reason about and attend to spatial information, highlighting its potential in aerial robotics applications.
Elena Wachtler, Antonella Barisic Kulas, Tamara Petrović et al.· International Conference on...· 0 citations
To address the inherent limitations of Vision-Language Models in long-tail object retrieval for autonomous driving, this paper proposes a Dual-Granularity Structured Scene Retrieval (DG-SSR) architecture. By decoupling text queries and visual features, we introduce a parameter-free mechanism that fuses local semantic scores with macro global context. An Adaptive Negative Injection (ANI) strategy and a Soft-NCE loss further enforce fine-grained alignment and mitigate color bias. Evaluated on a curated nuScenes dataset comprising 8,500 homogeneous street views and 1,128 combinatorial queries, our method achieves an mP@5 of 33.06% with a single-query latency of 0.812 ms, outperforming CLIP (16.48%) and BLIP (23.88%) by significant margins. Extensive ablation analysis demonstrates that optimal retrieval in complex scenes is achieved through a local-dominated architecture supplemented by minimal global context.
Nan Jiang, Tongxuan Xu, Yu-Jin Wang et al.· 2026 IEEE International Conf...· 0 citations
This work proposes ARIES-Mission2, a zero-shot Vision-Language-Action (VLA) framework that decouples visual-semantic perception from physical route optimization in low-altitude Unmanned Aerial Vehicle (UAV) mission generation and indicates that the TSP module maintains lower growth in computation time as the number of targets increases.
Junhao Wei, Yanxiao Li, Hao-Chen Li et al.· 0 citations
Global welfare often depends on the correct interpretation of aerial and satellite imagery. Acting on such imagery (mapping flooded ground, crop extent, or damaged infrastructure) demands pixel-level segmentation to ensure perfect class localization. Pretrained general foundation models, when applied directly, often miss important features and cannot always find all the classes belonging to a given scene, overlooking smaller objects that matter most. We use a single consumer-grade GPU running a vision-language model (VLM) to supply this missing guidance, improving segmentation while producing structured, auditable evidence that drives the result and can be inspected on its own. We fuse three approaches: the frozen foundation model that labels every pixel, and two queries to a VLM, one to choose the classes that matter, and one to locate the small objects the base model misses. Evaluating across four aerial datasets, we see consistent gains at each stage where the base model is competent.
Teresa DiMeola, Charles Walter, Hong Xiao· 0 citations
MoRAL (Multimodal Reasoning for Autonomous Language Models), a two-stage fine-tuning pipeline that teaches Cosmos-Reason2-2B to first read a physics-encoded Bird's Eye View (BEV) representation and then reason over it for driving decisions, establishes a reproducible foundation for compact, physics-grounded VLM reasoning on mobile edge platforms.
Ambarish Govindarajulu Kaliamurthi, Kai Liu· 0 citations
LiDAR-based semantic segmentation is a core perception module for autonomous vehicles and mobile robots. Despite the strong performance of recent state-of-the-art methods on standard benchmarks, existing evaluation protocols remain focused on clean, single-domain settings and fine-grained label taxonomies, leaving deployment readiness largely unassessed. Real-world systems must handle safety-critical label semantics, degraded sensing conditions, and cross-domain variability, yet no unified protocol currently addresses all three aspects together. In this paper, we propose a structured evaluation protocol that assesses the deployment readiness of LiDAR semantic segmentation models along three complementary dimensions: (i) coarse-label evaluation aligned with autonomous driving safety priorities, revealing how label granularity affects different methods; (ii) robustness under eight types of LiDAR corruptions designed to emulate real-world atmospheric, geometric, and sensor degradations; and (iii) domain generalization across datasets without adaptation. The evaluation includes inference speed measured on an embedded Jetson AGX Orin platform, directly reflecting deployment constraints. Our results show that fine-grained benchmark rankings do not always reflect safety-relevant performance, that all methods experience substantial degradation under corruptions with architecture-dependent robustness characteristics, and that current domain generalization remains insufficient for reliable deployment. These findings expose concrete gaps between benchmark performance and deployment readiness, and provide a reference protocol for more practically grounded evaluation of LiDAR semantic segmentation.
Samir Abou Haidar, Alexandre Chariot, Mehdi Darouich et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
