In this paper, we present a comprehensive dataset comprising 37.9 hours of sensor data collected from humanoid robots, including 18.3 hours of walking and 2,519 recorded falls. This extensive dataset is a valuable resource for various robotics and machine learning applications. Leveraging this data, we propose RePro-TCN, a Temporal Convolutional Network (TCN) enhanced with two novel extensions: Relaxed Loss Formulation and Progressive Forecasting. Predicting falls is a critical capability in humanoid robotics for implementing countermeasures such as lunging or stopping the walk. Thanks to the new dataset, we train RePro-TCN and demonstrate its superiority over previous approaches under real-world conditions that were previously unattainable.

Smart warehouses face rapid layout reconfigurations and frequent process adaptations, especially when using a Cyber-Physical Production Systems (CPPS) setup. These dynamic conditions introduce environmental uncertainty, making real-time navigation and obstacle avoidance a significant challenge for robot fleets. Limited sensing range and frequent occlusions further hinder local robot perception. Classic methods rely on centralized planning or vision-based systems, which struggle in low-visibility and cluttered environments. To address these gaps, we propose a graph-based collaborative perception network for the RoboFUSE (Framework for Unified Sensing and Exploration) system. Each robot operates onboard RoboFUSE with a dual-purpose waveform for sensing and communication (S&C) that emulates 6G Integrated Sensing and Communication (ISAC). This setup supports real-time data sharing across the fleet. On top of this platform, we develop RoboFUSE-Graph Neural Network (GNN), an uncertainty-aware GNN that fuses multi-robot radar data into a global spatial-semantic map. The model captures spatial relations and temporal dependencies using sliding window graphs. Experiments reveal an F1 score of 0.91 with a five-sliding window. The proposed approach enhances situational awareness, enables safe, scalable navigation, and represents a stride towards 6G-enabled smart warehouses.

This work proposes a novel image dataset for multi-camera multi-object tracking and a framework that allows users to generate similar datasets. The dataset, called Synth- Yard-MCMOT-l, is the first of its kind to be generated in a virtual environment with the main focus on the tracking of trucks in yard logistics environments. The dataset consists of a total of 12,008 images generated by eight different cameras. The images contain 44,232 bounding boxes and segmentation masks and 52 individual tracks. Additionally, we provide a ninth camera, which is used to generate unified ground-truth information for the whole scene from an orthographic, top-down perspective comparable to a bird's eye or map-view. The purpose of this dataset is to provide yard management systems with relevant data, which can be employed when aiming to determine the exact position of a truck and specifically identifying which gateway or designated parking spot it is located in. The purpose of the repository is to enable researches to create unique use-case-specific multi-camera tracking datasets with the included dataset-generation pipeline. Initial benchmarks for single-camera tracking demonstrate a mean identification F1 score score of 0.96 and a mean multiple object tracking accuracy score of 0.94, laying the baseline for computing world coordinates via multi-camera multi-object tracking.

This article presents a toolchain for event detection of pallets using IMUs and a novel data set called SPARL. Based on a logistical use case, two sensors are benchmarked in well-defined logistics scenarios. The data set contains the videos of two representative recordings from different perspectives and the raw data of the three used sensors. A random forest time series classification model is deployed for activity recognition. The preliminary results indicate that it is possible to recognize logistical activities based on sensor data.

This contribution explores the impact of synthetic training data usage and the prediction of material wear and aging in the context of re-identification. Different experimental setups and gallery set expanding strategies are tested, analyzing their impact on performance over time for aging re-identification subjects. Using a continuously updating gallery, we were able to increase our mean Rank-1 accuracy by 24 %, as material aging was taken into account step by step. In addition, using models trained with 10% artificial training data, Rank-1 accuracy could be increased by up to 13 %, in comparison to a model trained on only real-world data, significantly boosting generalized performance on hold-out data. Finally, this work introduces a novel, open-source re-identification dataset, pallet-block-2696. This dataset contains 2,696 images of Euro pallets, taken over a period of 4 months. During this time, natural aging processes occurred and some of the pallets were damaged during their usage. These wear and tear processes significantly changed the appearance of the pallets, providing a dataset that can be used to generate synthetically aged pallets or other wooden materials.

The growth of parcel shipments for end-customer delivery led to the emergence of robotic sorting systems, which offer improved flexibility and cost-effectiveness compared to conveyor-based sorting systems. Controlling large fleets of mobile robots for sorting tasks remains a challenge, as it requires solving a complex multi-agent path finding problem. One possible solution is the recently introduced rolling horizon collision resolution framework, which is actively discussed in the literature. However, its configuration to suit robotic sorting systems concerning real-time capabilities and throughput is still the object of ongoing investigation. To address this research gap, this study investigates the application of the rolling horizon collision resolution framework on two types of robotic sorting systems and a parameter study is conducted to assess its potential in terms of real-time capability and sorting performance.

In recent years, the Robotic Mobile Fulfillment System has been established as a new goods-to-person storage system, which particularly addresses the needs of e-commerce. In these systems, a fleet of mobile robots carries inventory pods (mobile racks) between picking stations and storage locations, a task that requires efficient path planning for potentially hundreds of robots. Therefore, this task can be considered an instance of the Multi-Agent Path Finding problem, where the goal is to find collision-free and goal-reaching paths for a set of agents. Previous publications addressing Multi-Agent Path Finding for Robotic Mobile Fulfillment Systems use oversimplified goal-assignment schemes and use-cases. To address these issues, we present an adapted version of the Multi-Agent Path Finding problem that mimics the goal assignment scheme of real-world picking systems and we introduce three industry-inspired use-cases. Finally, using the Rolling Horizon Collision Resolution framework, we apply three state-of-the-art solvers for Multi-Agent Path Finding problems to our use-cases. Our preliminary results indicate that two of the three solvers are suitable for usage in Robotic Mobile Fulfilment systems.

Today, many industrial tasks are not automated and still require human intervention. One of these tasks is the unloading of oversea containers. After the end of transportation to the sorting center, the containers must be unloaded manually for further sending the parcels to the recipients. A robot-based automatic unloading of containers was therefore researched. However, the promising results of the system developed in these projects could not be commercialized due to problems with its reliability. Mechanical, algorithmic or other limitations are possible causes of the observed errors. To analyze errors, it is necessary to evaluate the results of the robot’s work without complicating the existing system by adding new sensors to it. This paper presents a reference system based on machine learning to evaluate the robotics grasps of parcels. It analyzes two states of the container: before and after picking up one box. The states are represented as a point cloud received from a laser scanner. The proposed system evaluates the success of transferring a box from an overseas container to the sorting line by supervised learning using convolutional neural networks (CNN) and manual labeling of the data. The process of obtaining a working model using a hyperband model search with a maximum classification error of 3.9 % is also described.