The complexities of decision-making in drone airlines prove to be pivotal and challenging as the dynamic environment introduces variability and many decisions are conventionally static. This paper introduces an advanced decision-making system designed for the multifaceted landscape of drone applications. Our proposed system addresses various aspects, including drone assignment, safety zone sizing, priority determination, and more. The scoring model enhances adaptability in real-time scenarios, particularly highlighted by the dynamic adjustment. Based on the scenario concerning the definition of the safety zone, we have successfully applied this method and evaluated all potential scores. The user-friendly and intuitive configuration further augments the system’s accessibility, facilitating efficient deployment. In essence, the proposed system stands as an innovative approach with decision-making paradigms in the dynamic landscape of drone operations.

Street delivery faces significant challenges due to outdated road infrastructure, which was not designed to handle current vehicle volumes, leading to congestion and inefficiencies, especially in last-mile delivery. Integrating drones into the delivery system offers a promising solution by bypassing congested roads, thereby enhancing delivery speed and reducing infrastructure strain. However, optimizing this multimodal delivery system is complex and data-driven, with real-world data often being costly and restricted. To address this, we propose a synthetic dataset generator that creates diverse and realistic delivery scenarios, incorporating environmental variables, customer profiles, and vehicle characteristics. The key contribution of our work is the development of a dynamic generator for multiple optimization problems with diverse complexities or even combinations of optimization problems. This generator allows for the incorporation of real-world factors such as traffic congestion and synthetically generated factors such as wind conditions and communication constraints, as well as others. The primary objective is to establish a foundation for creating benchmark scenarios that enable the comparison of existing and new approaches. We evaluate the generated dataset by applying it to three optimization problems, including facility location, vehicle routing, and path planning, using different techniques to demonstrate the dataset’s effectiveness and operational viability.

Deep networks have been successfully applied to industrial applications for clean unimodal data (e.g., sensors, images, or audio). Leveraging multimodal data is a common approach to enhance performance, guided by the principle that a larger quantity of data leads to improvement. However, performance may decline considerably if corruption in the data is present (e.g., noise, blur, failure). Although researchers have explored various data augmentation methods to improve the generalization capacity, these methods are not adapted for industrial settings. The primary distinction is that current augmentation methods are designed to enhance model generalization capabilities and not realistically simulate real-world industry scenarios. In this paper, we present industry-related augmentation methods for temporal and spatial data for multimodal fusion with deep neural networks. Our methods are specifically designed to encourage modality collaboration and reinforce generalization capability. The impact of the proposed data extension strategy to train multimodal fusion models is assessed on a synthetic dataset from an industrial UR5 robot with varying degrees of imbalance. In our study, we analyze different combinations of methods and evaluate their performance. Through these experiments, we are able to identify the challenges in multimodal fusion with deep learning models in an industrial setting.

Deep networks have been successfully applied to unsupervised feature learning and supervised classification and regression for unimodal data (e.g., sensors, images, or audio). Multimodal data is often used to improve the performance of networks according to the slogan: the more, the better. Limited research is available to compensate for corrupted signals from multimodal approaches. In this work, we propose a novel regularization method for deep networks to learn features over multiple modalities designed to compensate for relative sensor weaknesses, such as sensor malfunction, inaccuracy, restricted spatial coverage, and uncertainty. We have used a special augmentation strategy for image and time series modalities to enhance the dataset of underrepresented industrial failure cases. The primary objective is to prevent these cases from negatively impacting the model's predictions. Our approach involves incorporating a fuzzy regularizer that can modify the intensity of activations depending on the quality of the signal, enabling disturbances from various modalities to be identified based on the activations. Our experiments on a simulated Universal Robots UR5 dataset demonstrate the effectiveness of our proposed regularization in increasing the model's stability, accuracy, and generalization to uncertainties and failures in the input signals. By incorporating fuzzy regularization in deep fusion models, their efficiency on complex tasks can be improved while reducing the complexity of the architectures.

Optimizing industrial processes is crucial, as unplanned downtime in high-capacity production lines incurs significant costs due to energy, labor, and material losses. While predictive maintenance helps mitigate predictable wear and aging, it falls short in addressing random, unforeseen failures that often cause unexpected downtime. In this study, we introduce the Multimodal Concept Fusion Architecture, a novel deep learning framework designed to robustly integrate heterogeneous and multimodal corrupted representations. The proposed model focuses on fusing latent features extracted from structurally distinct modalities by leveraging a learned fusion concept that adaptively balances joint and contrastive information. Additionally, this approach inherently enables anomaly detection and offers deeper opportunities for interpretation. By leveraging concept-guided fusion, it becomes possible to trace decision pathways back to modality-specific contributions, increasing transparency and trust in system behavior. We prove the practicality and robustness of our method by performing experiments on three synthetically constructed yet industrially realistic multimodal datasets, including various sensor failures through targeted data augmentations.