Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for controlling an agent interacting with an environment using a Transformer neural network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for controlling an agent interacting with an environment. In one aspect, a method comprises: receiving a user text sequence that characterizes a robotic task to be performed by the agent in the environment; converting the user text sequence into an input text sequence in a standardized format for a visual language understanding task; obtaining an observation image characterizing a state of the environment; processing (i) the input text sequence in the standardized format and (ii) the observation image using a policy neural network to generate a policy output that defines an action to be performed by the agent in response to the observation image; selecting an action to be performed by the agent using the policy output; and causing the agent to perform the selected action.

Abstract: In various examples, a computing system can receive a caption regarding an image. The computing system can generate, responsive to detecting that the caption includes at least a threshold amount of text, based at least on applying text recognition to the image, text data represented in the image. The computing system can cause at least one language model to generate an annotation for the image based at least on the caption, the text data, an example query for the language model to process, and an example response corresponding to the example query.

Abstract: Various implementations disclosed herein include devices, systems, and methods that use frames of sensor data to create key frames that are used to provide a 3D representation of a physical environment (e.g., a 3D mesh, planar representations, object bounding boxes, a combo of these, etc.). Some implementations use key frames (e.g., bundles of 2D image data deemed to be representative or useful for reconstruction/3D representation generation). Such key frames may be created by evaluating multiple frames of sensor data and determining whether to convert a given frame into a key frame or fuse it, e.g., fuse it with a nearest existing key frame. Such fusing may reduce loss of information which would occur if frames not selected as key frames are simply discarded.

Abstract: Video-based multimodal large language models that include an image encoder configured to transform frames from a video into visual tokens, a multimodal large language model configured to receive the visual tokens, and a temporal module interposed between the image encoder and the multimodal large language model.

Abstract: This disclosure relates to a power converter, a slope signal generator and a slope signal generation method thereof. The power converter comprises an inductor and a switch transistor coupled between the input terminal and the output terminal. The slope signal generator comprises: a proportion transformation circuit, for performing proportion transformation on the current sense signal of the inductor to obtain a replicated signal; and an offset compensation circuit, for performing offset compensation on the replicated signal to obtain a slope signal, wherein the slope change of the slope signal characterizes the inductance value change of the inductor. The slope signal generator of the power converter can adaptively adjust the slope of the slope signal during the switching cycle, dynamically compensate for the fluctuation of the inductance value, improve the effect of suppressing subharmonic oscillation, thus enhancing the input and output voltage range and stability of the power converter.

Abstract: The present invention provides a solid preparation containing (7aS,2?S)-2-oxo-clopidogrel and a preparation method therefor. The solid preparation is an oral tablet containing (7aS,2?S)-2-oxo-clopidogrel and pharmaceutically acceptable excipients, the excipients comprising a filler, a disintegrant, a binder, and a lubricant.

Abstract: Approaches are described for defect detection and analysis in semiconductor substrate (e.g., wafer/reticle) inspection using neural networks and computational models. Disclosed approaches can process images of a semiconductor substrate to identify expected features, such as alignment marks, interconnects, or functional geometries, which represent the intended characteristics of the patterned wafer/reticle's design. These expected features may be provided in various forms, including text-based prompts, CAD-derived reference images, or other design-related data. Such analysis can involve comparing input images with these benchmarks to detect deviations, such as missing features, misalignments, overlapping patterns, or geometric inconsistencies. An example system can allow for analysis of individual images, sub-regions, or multiple images captured across different fabrication stages. Real-time data streams from manufacturing equipment may also be processed for continuous monitoring.

Abstract: A transmitter inverts polarities of some symbols in a to-be-sent signal based on a target frequency, so that power of a to-be-sent signal at the target frequency is suppressed. The transmitter may flexibly adjust the power spectrum of the to-be-sent signal based on an actual frequency of interference or noise, so that the to-be-sent signal does not overlap the interference or the noise in frequency, and a receiver can effectively filter out the interference or the noise and cause no impairment to the signal from the transmitter. The transmitter further sends identification information to the receiver, so that the receiver performs inverse transformation of power spectrum adjustment on a received signal based on the identification information, to restore the original signal that is before the transmitter performs power spectrum adjustment.

Abstract: Embodiments of this application disclose a signal processing method and a related apparatus. The signal processing method may be applied to an optical communication scenario. The method in embodiments of this application includes: a transmitter inverts polarities of some symbols in a first modulated signal to obtain a second modulated signal, so that an absolute value of a sum of values of all symbols in the second modulated signal is less than an absolute value of a sum of values of all symbols in the first modulated signal. In this way, power of a low-frequency part in the second modulated signal is suppressed. In addition, the receiver performs inverse transformation of power spectrum adjustment on the received signal based on identification information, to restore the original signal that is before the transmitter performs power spectrum adjustment.

Abstract: An apparatus comprising an interface and a processor. The interface may be configured to receive pixel data of an environment. The processor may be configured to process the pixel data arranged as video frames, generate a luminance distribution map of the video frames in response to a low-pass filter operation, determine a plurality of defogging intensity weights for the luminance distribution map, perform adaptive smoothing to each of the plurality of defogging intensity weights, and generate defogged video frames in response to the video frames and the plurality of defogging intensity weights with the adaptive smoothing. The plurality of defogging intensity weights may each correspond to one of a plurality of luminance intervals of the luminance distribution map. The adaptive smoothing may be configured to prevent brightness differences in the defogged video frames.

Abstract: Apparatuses, systems, and techniques to obtain one or more captions for a video using machine learning. In at least one embodiment, at least one machine learning process is used to generate at least one output caption using at least one image-level caption, at least one video-level caption, and/or at least one motion caption. In at least one embodiment, the video-level caption(s) is/are generated by one or more second machine learning processes using the video, and the image-level caption(s) is/are generated by one or more third machine learning processes using one or more images sampled from the video.

Abstract: An end mill with a gradient edge geometry and a grinding method thereof relate to the technical field of metal cutting. Through multi-dimensional data acquisition, the grinding method comprehensively evaluates a cutting load, a cutting temperature and a cutting vibration amplitude of the tool, and overcomes limitations of the conventional method. By setting a lifetime evaluation criterion and an alert threshold, the grinding method forms a scientific evaluation system, improving accuracy of the evaluation and lifetime of the tool. By combining a preliminary adjustment strategy with a fine adjustment strategy, the grinding method improves the adaptability of the tool in complex conditions, ensuring that the tool works in an optimal state all the time.

Abstract: A solid heat carrier catalyst for thermal desorption of organic matter-contaminated soil and a method for preparing the same. A hollow alumina ball prepared by 3D printing is taken as a solid heat carrier, copper-nickel-vanadium composite oxide is taken as a catalytic active component, and vinyltriethoxysilane is taken as a masking agent. The ball has a diameter of 30 mm to 60 mm and a thickness of 1 mm to 2 mm. An outer surface of the ball is masked with the vinyltriethoxysilane; then the ball is pierced to make an inner surface thereof connected with the outside through channels; the ball is then immersed in a catalytic active component precursor solution; and finally, drying and calcination are performed to obtain the solid heat carrier catalyst for thermal desorption of organic matter-contaminated soil. This product is widely applicable to the field of thermal desorption of organic contaminants of soil.

Abstract: Batteries having convoluted, freeform shapes. These batteries can have curved and twisted surfaces that are freeform and do not have a constant radius or degree of rotation. A stack of battery layers can undergo a two-step heat press. The first heat press can be applied to the battery layers in a limited, low-stress region. This heat-press step can help to reduce delamination during a second heat-press step. In this second heat-press step, some or all of the battery layers can be pressed into a freeform shape and heat can be applied. The battery layers can then be placed in an enclosure.

Abstract: A laser is provided, which includes a frame, a substrate, heat sinks, light-emitting chips, and protective devices. The frame and the heat sinks are fixed to the substrate. The heat sinks, the light-emitting chips, and the protective devices are located inside the frame. The light-emitting chips and the protective devices are fixed to the heat sinks, and the light-emitting chip and the corresponding protective device have a spacing therebetween in a length direction of the heat sink.

Abstract: Apparatuses, systems, and techniques are disclosed for controlling a robot to execute a task. In at least one embodiment, a current image of the robot in an environment and a text describing the task are obtained. A future image of the robot in the environment is predicted based on the current image and the text. Subsequently, one or more actions are predicted based on the current image, the future image, and the text. The one or more actions can move the robot from a first state corresponding to the current image to a second state corresponding to the future image. The robot executes the sequence of actions to move in the environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for controlling an agent interacting with an environment. In one aspect, a method comprises: receiving an observation that characterizes the environment; receiving a conditioning input that characterizes a task to be performed by the agent in the environment; for each of a plurality of sub-regions of the observation, generating an observation patch embedding of the sub-region; generating a conditioning input embedding of the conditioning input; processing the observation patch embeddings and the conditioning input embedding to generate a policy output that defines an action to be performed by the agent in response to the observation, wherein the processing comprises applying a linear attention mechanism over the observation patch embeddings and the conditioning input embedding; selecting an action to be performed by the agent using the policy output; and causing the agent to perform the selected action.