A sensor transformation attention network (STAN) model including sensors configured to collect input signals, attention modules configured to calculate attention scores of feature vectors corresponding to the input signals, a merge module configured to calculate attention values of the attention scores, and generate a merged transformation vector based on the attention values and the feature vectors, and a task-specific module configured to classify the merged transformation vector is provided.

Videos captured in low light conditions can be processed in order to identify an activity being performed in the video. The processing may use both the video and audio streams for identifying the activity in the low light video. The video portion is processed to generate a darkness-aware feature which may be used to modulate the features generated from the audio and video features. The audio features may be used to generate a video attention feature and the video features may be used to generate an audio attention feature. The audio and video attention features may also be used in modulating the audio video features. The modulated audio and video features may be used to predict an activity occurring in the video.

The present teaching relates to system, method, medium for in-situ perception in an autonomous driving vehicle. A plurality of types of sensor data acquired continuously by a plurality of types of sensors deployed on the vehicle are first received, where the plurality of types of sensor data provide information about surrounding of the vehicle. Based on at least one model, one or more items are tracked from a first of the plurality of types of sensor data acquired by one or more of a first type of the plurality of types of sensors, wherein the one or more items appear in the surrounding of the vehicle. At least some of the one or more items are then automatically labeled on-the-fly via either cross modality validation or cross temporal validation of the one or more items and are used to locally adapt, on-the-fly, the at least one model in the vehicle.

A device for recognizing a fingerprint, includes: a fingerprint sensor; at least two moisture detection electrodes disposed within a preset range of the fingerprint sensor; and a processing module coupled to the fingerprint sensor and the at least two moisture detection electrodes. The fingerprint sensor is configured to output a fingerprint signal to the processing module when a user touches the fingerprint sensor and the at least two moisture detection electrodes with a finger. The processing module is configured to acquire a characteristic value which is positively related to an impedance between the at least two moisture detection electrodes when the user touches the fingerprint sensor and the at least two moisture detection electrodes with the finger; determine a fingerprint recognition parameter which matches the characteristic value; and perform fingerprint recognition according to the determined fingerprint recognition parameter and the fingerprint signal.

A method comprising: obtaining a three-dimensional (3D) point cloud about an object; obtaining binary feature descriptors for feature points in a 2D image about the object; assigning a plurality of index values for each feature point as multiple bits of the corresponding binary feature descriptor; storing the binary feature descriptor in a table entry of a plurality of hash key tables of a database image; obtaining query binary feature descriptors for feature points in a query image; matching the query binary feature descriptors to the binary feature descriptors of the database image; reselecting one bit of the hash key of the matched database image; and re-indexing the feature points in the table entries of the hash key table of the database image.

An apparatus for performing continuous actions includes a memory storing instructions, and a processor configured to execute the instructions to obtain a first action of an agent, based on a current state of the agent, using a cross-entropy guided policy (CGP) neural network, and control to perform the obtained first action. The CGP neural network is trained using a cross-entropy method (CEM) policy neural network for obtaining a second action of the agent based on an input state of the agent, and the CEM policy neural network is trained using a CEM and trained separately from the training of the CGP neural network.

A device implementing a system for estimating device location includes at least one processor configured to receive a first estimated position of the device at a first time. The at least one processor is further configured to capture, using an image sensor of the device, images during a time period defined by the first time and a second time, and determine, based on the images, a second estimated position of the device, the second estimated position being relative to the first estimated position. The at least one processor is further configured to receive a third estimated position of the device at the second time, and estimate a location of the device based on the second estimated position and the third estimated position.

A computing system accesses one or more data sources to determine maintenance optimization data associated with an asset within a set of assets. The maintenance optimization data may include one or more of: upcoming maintenance events for the asset, such as may be predicted based on analysis of historical maintenance information of the asset, a time series of predicted value of the asset over a time period around the upcoming maintenance event, such as within a few days or hours of the maintenance event, and/or a recommended window of time to initiate and/or perform upcoming maintenance events, which may be based on a combination of the expected upcoming maintenance events, and the time series of predicted value of the particular asset, for example.

A control method includes obtaining feature data using at least one sensor, the feature data being acquired by the terminal using the at least one sensor, generating an action instruction based on the feature data and a decision-making mechanism of the terminal, and executing the action instruction. In this application, various aspects of feature data are acquired using a plurality of sensors, data analysis is performed on the feature data, and a corresponding action instruction is then generated based on a corresponding decision-making mechanism to implement interactive control.

Described herein are assistant robots that anticipate needs of one or more people (or animals). The assistant robots may recognize a current activity, knowledge of the person's routines, and contextual information. As such, the assistant robots can provide or offer to provide appropriate robotic assistance. The assistant robots can learn users' habits or be provided with knowledge regarding humans in its environment. The assistant robots develop a schedule and contextual understanding of the persons' behavior and needs. The assistant robots may interact, understand, and communicate with people before, during, or after providing assistance. The robot can combine gesture, clothing, emotional aspect, time, pose recognition, action recognition, and other observational data to understand people's medical condition, current activity, and future intended activities and intents.