Robert Bosch

Es wird ein Batterieadapter (100) für ein Elektrowerkzeug (200) mit einem Kontaktierungselement (120) zur elektrischen Kontaktierung mit dem Elektrowerkzeug (200), mit einem Führungselement (140) zur Führung in einer Führungsaufnahme (202) des Elektrowerkzeugs (200) und mit einem Verriegelungselement (160) zur Verriegelung mit einem Gehäuse (206) des Elektrowerkzeugs (200), offenbart.Es wird vorgeschlagen, dass der Batterieadapter (100) ein Verschlusselement (180) aufweist, das dazu ausgebildet ist, das Verriegelungselement (160) zu verschließen., offenbart.

Synchronized camera images and radar images are captured from a scene. A shared spatial representation of the scene is generated by encoding spatial features into a spatial hash table of a shared geometry encoder. The shared spatial representation is decoded using a geometry decoder to produce camera occupancy and radar occupancy values. Camera and radar poses are refined through an optimization process that estimates per-frame pose and velocity offsets. A normal multi-layer perceptron (MLP) is used to predict surface normals at spatial locations of the scene based on the shared spatial representation, and one or more bidirectional reflectance distribution function (BRDF) bases are used to model radar reflectance. Camera density and radar density values are generated using the BRDF bases and the predicted surface normals. Predicted camera images and radar images are rendered. The shared geometry encoder is optimized based on a multimodal loss function.

A computer-implemented technique for outputting a 3D feature volume of a scene is provided. A method comprises a step of receiving (S102) multi-view 2D image data of a scene. Features are extracted (S105) at multiple scales of spatial resolution. At an input layer of a 3D backbone network (307), 3D point cloud data of the scene are received (S108). The 3D backbone network (307) comprises an encoder (309-E) and a decoder (309-D) with connections (410) at a plurality of scales of spatial resolution for extracting features from the 3D point cloud data. At multiple layers of the decoder (309-D) the extracted (S105) features are injected (S110) with corresponding scales of spatial resolution. A 3D feature volume of the scene based on the received (S108) 3D point cloud data and on the injected (S110) features is output (S112) at an output layer of the 3D backbone network (307).

Methods 4000, 5000 for providing and using a scalable machine learning model 20 for performing application task(s) 121, 122, 123 are provided. A primary machine learning model 10, comprising a plurality of model parameters 11, 12, 13, is trained on training data. A secondary machine learning model 21, a subset of the primary machine learning model, is provided to serve as a substitute. The primary and secondary machine learning models are jointly trained by computing outputs for the one or more application tasks of the primary machine learning model for the training data, and computing, based on the outputs and the training data, a first loss for the primary machine learning model; analogously computing a second loss for the secondary machine learning model; combining the first and second losses; and updating the plurality of model parameters of the primary and thereby those of the second machine learning model accordingly. The jointly trained models are then provided as scalable machine learning model.

Computer-implemented method (900) for determining an uncertainty (u) of a sensor signal (x1,x2 ,xM) synthesized by a generative machine learning system (61) with respect to how likely it is to observe the sensor signal (x1 ,x2 ,xM) in physical reality comprising the steps of: • Obtaining (901) a noise sample (z); • Performing (902) Bayesian Inference on the generative machine learning (61) system using the noise sample (z) as input to the generative machine learning system (61) thereby determining a posterior predictive distribution of the sensor signal that would have been synthesized by the generative machine learning system (61) using the noise sample (z) as input to the generative machine learning system (61); • Providing (903) a measure of variability of the posterior predictive distribution as uncertainty with respect to a sensor signal (x1,x2 ,xM) synthesized from the noise sample (z), wherein the method is characterized in the posterior predictive distribution characterizing a distribution of latent features (e1 ,e2 ,eM) of the sensor signal (x1,x2 ,xM) that would have been generated using the noise sample (z) as input.

A method for training a machine learning, ML, model for generating a joint particle-based representation (510) of a scene from multi-modality sensor data comprises receiving sensor data (502; 5012; 522) acquired by a set of optical sensors of different sensor modality, representing the same scene. A joint particle-based representation (510) of the scene covered by the optical sensors is generated. Intersections between rays emanating from a sensor location with the joint particle-based representation (510) are determined, based on which an observation of the scene is rendered (508; 518; 528). A sensor-specific loss function value is determined by comparing the rendered (508; 518; 528) observation with an observation rendered based on the received sensor data (502; 5012; 522). The ML model is trained by determining the intersections, rendering the observation and determining the value of the sensor-specific loss function for every optical sensor, and by optimizing a combination (520) of the sensor-specific loss functions.

The present invention relates to a method of operating a solenoid valve (1), said solenoid valve being configured to regulate a flow of fluid along a conduit (10) along which said solenoid valve is positioned, said method comprising the following steps:a. Receiving a command to bring said solenoid valve from a closed position, at which said solenoid valve prevents the flow of fluid along said conduit, to an open position of said valve, in which said open position allows at least partial passage of fluid along said conduit through said solenoid valve;b. Providing a load signal, said load signal being configured to directly or indirectly describe a pressure along said conduit at said solenoid valve;c. Providing said solenoid valve with a current signal to enable said solenoid valve to assume said open position, wherein said current signal comprises a current peak having an amplitude, wherein said amplitude of said peak depends on said load signal provided in said step b.

Verfahren (100) zum Trainieren eines neuronalen Netzwerks (2), das eine oder mehrere Kenngrößen (3) einer vorgegebenen Szenerie (1) vorhersagt, mit den Schritten: • es wird eine Menge von Trainings-Beispielen (6) bereitgestellt (110), die jeweils mindestens Kombinationen aus Messorten (4) einerseits und Referenz-Beobachtungen (4a) der Szenerie (1) andererseits enthalten; • für jedes Trainings-Beispiel (6) wird der jeweilige Messort (4) dem neuronalen Netzwerk (2) zugeführt (120); • die so erhaltenen Kenngrößen (3) werden mit einem vorgegebenen Rendering-Modell (7) in eine unter idealisierten Bedingungen ausgehend vom jeweiligen Messort (4) zu erwartende Ideal-Beobachtung (4a*) der Szenerie (1) übersetzt (130); • die Ideal-Beobachtung (4a*) wird mit einem Messmodell (8) in eine unter realistischen Bedingungen ausgehend vom jeweiligen Messort (4) zu erwartende Real-Beobachtung (4a**) der Szenerie (1) übersetzt (140); • die Real-Beobachtungen (4a**) werden mit den Referenz-Beobachtungen (4a) für das jeweilige Trainings-Beispiel (6) verglichen (150); • eine bei diesem Vergleich festgestellte Abweichung (Δ) wird mit einer vorgegebenen Kostenfunktion (9) bewertet (160); und • Parameter (2a), die das Verhalten des neuronalen Netzwerks (2) charakterisieren, werden auf das Ziel optimiert (170), bei weiterer Verarbeitung von Trainings-Beispielen (6) die Bewertung (9a) durch die Kostenfunktion (9) zu verbessern.

Die vorliegende Erfindung betrifft ein Verfahren (10) zum Betreiben eines Sensorsystems (50) eines Elektrofahrrads (100), wobei das Sensorsystem (50) einen Drehmomentsensor (55) umfasst, welcher als ein magnetoelastischer Sensor ausgebildet ist, und welcher eingerichtet ist, um basierend auf einer Erfassung von Änderungen eines Magnetfeldes ein Drehmoment an einer Welle (101) zu ermitteln, umfassend die Schritt: Erfassen (1) einer homogenen magnetischen Flussdichte mittels des Drehmomentsensors (55), Ermitteln (2) eines Drehmoments an der Welle (101) mittels des Drehmomentsensors (55), und Erkennen (3) eines Drehmomentfehlers des ermittelten Drehmoments, wenn die erfasste homogene magnetische Flussdichte größer oder gleich einem vorbestimmten ersten Schwellwert ist.

Die Erfindung betrifft ein Rollenmodul (30) für einen Rollenförderer (10) mit Königswellenantrieb. Erfindungsgemäß ist ein Grundkörper eines Rollenmoduls (30) wannenartig und einstückig ausgebildet.