罗伯特·博世公司

Especially in urban traffic, the field of view of road users is often occluded by buildings, vegetation, or other road users. Thus, conservative driving decisions have to be made, which allow one to always react appropriately and obey traffic rules. Intelligent infrastructure along the roadside, data fusion and cooperative planning services on edge computing servers, and vehicle-to-anything communication can help overcome this limitation and-thus-improve traffic efficiency while retaining safety for all road users. This article summarizes the architecture of such a system, prototyped in Ulm, Germany. Further, it presents results from tests involving connected automated and non-automated vehicles and connected vulnerable road users in regular non-connected traffic, realizing different cooperative maneuvers. Besonders im Stadtverkehr behindern Geb & auml;ude, Vegetation oder andere Verkehrsteilnehmer die Sicht der Verkehrsteilnehmer. Dies erfordert eine entsprechend vorsichtigere Fahrweise, um in jeder Situation angemessen reagieren zu k & ouml;nnen und dabei stets die Verkehrsregeln einzuhalten. Intelligente Infrastruktur entlang der Stra beta e, Datenfusion und kooperative Planungsdienste k & ouml;nnen in Kombination mit vernetzten Fahrzeugen dazu beitragen, diese Einschr & auml;nkung zu & uuml;berwinden und damit die Effizienz des Verkehrsflusses zu verbessern, w & auml;hrend die Sicherheit der Verkehrsteilnehmer weiter gew & auml;hrleistet wird. Dieser Beitrag gibt einen & Uuml;berblick & uuml;ber die Architektur und Funktionsweise einer solchen intelligenten Infrastruktur, die prototypisch in Ulm realisiert wurde. Dar & uuml;ber hinaus werden Ergebnisse von Tests verschiedener kooperativer Man & ouml;ver mit vernetzten, automatisierten Fahrzeugen und vernetzten verletzlichen Verkehrsteilnehmern im ansonsten unvernetzten, realen Stra beta enverkehr vorgestellt.

Vision foundation models (VFMs) trained on large-scale image datasets provide high-quality features that have significantly advanced 2D visual recognition. However, their potential in 3D scene segmentation remains largely untapped, despite the common availability of 2D images alongside 3D point cloud datasets. While significant research has been dedicated to 2D-3D fusion, recent state-of-the-art 3D methods predominantly focus on 3D data, leaving the integration of VFMs into 3D models underexplored. In this work, we challenge this trend by introducing DITR, a generally applicable approach that extracts 2D foundation model features, projects them to 3D, and finally injects them into a 3D point cloud segmentation model. DITR achieves state-of-the-art results on both indoor and outdoor 3D semantic segmentation benchmarks. To enable the use of VFMs even when images are unavailable during inference, we additionally propose to pretrain 3D models by distilling 2D foundation models. By initializing the 3D backbone with knowledge distilled from 2D VFMs, we create a strong basis for downstream 3D segmentation tasks, ultimately boosting performance across various datasets.

Hydraulic excavators remain among the least automated construction machines due to nonlinear hydraulics, varying environment interactions, and strict real-time requirements. We present a physics-informed online learning framework that provides a standardized velocity-level control interface, abstracting machine-specific hydraulics and enabling integration with higher-level automation. A physics-motivated base model is augmented with an online residual learner to compensate nonlinearities in real time. Experiments on a 12-ton excavator in leveling, grading, and a demanding 4-point cycle show substantial improvements over the base model alone: maximum absolute tracking error is reduced by up to 63 % in leveling and 70 % in the 4-point cycle, while RMSE is reduced by up to 54 % and 69 %. Hydraulische Bagger z & auml;hlen nach wie vor zu den am wenigsten automatisierten Baumaschinen. Gr & uuml;nde hierf & uuml;r sind die Nichtlinearit & auml;ten der Hydraulik, variierende Interaktionen mit der Umgebung sowie strenge Echtzeitanforderungen. Wir stellen ein physik-informiertes Online-Lernframework vor, das eine standardisierte Regelungsschnittstelle auf Geschwindigkeitsebene bereitstellt. Dadurch werden maschinenspezifische hydraulische Eigenschaften abstrahiert und die Integration in & uuml;berlagerte Automatisierungsfunktionen erm & ouml;glicht. Ein physikalisch motiviertes Basismodell wird um einen online trainierten Residual-Lerner erweitert, der Nichtlinearit & auml;ten in Echtzeit kompensiert. Experimente mit einem 12-Tonnen-Bagger bei Planier- und Gradierungsaufgaben sowie in einem anspruchsvollen 4-Punkt-Zyklus zeigen deutliche Verbesserungen gegen & uuml;ber dem alleinigen Basismodell: Der maximale absolute Regelfehler wird beim Planieren um bis zu 63 % und im 4-Punkt-Zyklus um bis zu 70 % reduziert, w & auml;hrend der RMSE um bis zu 54 % bzw. 69 % sinkt.

The availability of large, high-quality emotional speech databases is essential for advancing speech emotion recognition (SER) in real-world scenarios. However, many existing databases face limitations in size, emotional balance, and speaker diversity. This study describes the MSP-Podcast corpus, summarizing our ten-year effort. The corpus consists of over 400 hours of diverse audio samples from various audio-sharing websites, all of which have Common Licenses that permit the distribution of the corpus. We annotate the corpus with rich emotional labels, including primary (single dominant emotion) and secondary (multiple emotions perceived in the audio) emotional categories, as well as emotional attributes for valence, arousal, and dominance. At least five raters annotate these emotional labels. The corpus also has speaker identification for most samples, and human transcriptions of the lexical content of the sentences for the entire corpus. The data collection protocol includes a machine learning-driven pipeline for selecting emotionally diverse recordings, ensuring a balanced and varied representation of emotions across speakers and environments. The resulting database provides a comprehensive, high-quality resource, better suited for advancing SER systems in practical, real-world scenarios.

Large Language Models (LLMs) training is prohibitively expensive, driving interest in low-precision fully-quantized training (FQT). While novel 4-bit formats like NVFP4 offer substantial efficiency gains, achieving near-lossless training at such low precision remains challenging. We introduce TetraJet-v2, an end-to-end 4-bit FQT method that leverages NVFP4 for activations, weights, and gradients in all linear layers. We identify two critical issues hindering low-precision LLM training: weight oscillation and outliers. To address these, we propose: 1) an unbiased double-block quantization method for NVFP4 linear layers with practically optimal convergence in LLM training, 2) OsciReset, the first effective algorithm to suppress LLMs' weight oscillation bottleneck, and 3) OutControl, a mix-precision algorithm to retain outlier accuracy. TetraJet-v2 outperforms prior methods on FP4 pre-training for LLMs across models up to 370M parameters trained up to 212B tokens, reducing the performance gap to BF16 by an average of 51.3