Abstract: A food preparation system and method include: triggering image capturing of a camera to obtain one or more images of a food support platform while the food support platform supports a first food item; performing ingredient recognition for the first food item, including: classifying a feature tensor of a respective image in a general classifier to identify one or more first-level food ingredient categories corresponding to the first food item; and classifying the feature tensor of the respective image in a respective detailed classifier corresponding to each of the one or more first-level food ingredient categories to identify a corresponding second-level food ingredient category corresponding to the first food item, wherein the second-level food ingredient category is a sub-category of said each first-level food ingredient category; and, adjusting one or more heating units for heating the first food item in accordance with the ingredient recognition that has been performed.

Abstract: System and method for controlling a home assistant device include: receiving an audio input; performing speech-to-text conversion on the audio input to obtain a textual string; searching a name list of multiple home devices to determine whether the textual string includes a respective alias of at least one home device in the name list; in accordance with a determination that the textual string includes a first alias corresponding to a first home device of a first device type, generating a first actionable intent to operate the first home device of the first device type; and in accordance with a determination that the textual string includes a second alias corresponding to a second home device of the first device type that is distinct from the first home device, generating a second actionable intent to operate the second home device of the first device type.

Abstract: The disclosed computer-implemented method for artificial-intelligence-based automated surface inspection can include receiving customer data, a request for a targeted model, and compensation for the requested targeted model. The compensation can include an agreement to contribute the customer data and/or targeted model to be available for other third-party entities. The method can also include retrieving the pre-trained model from a pre-trained model pool. The pre-trained model can be related to objects in a second industry. The method can include generating the targeted model from the pre-trained model and the customer data. The targeted model can be related to mapping sensor data to surface anomalies. The method can also include providing the targeted model to the third-party entity. The method can further include updating a distributed blockchain structure to include the at least one of the customer data and the targeted model.

Abstract: A method, a system, and a computer readable recording medium are disclosed for performing object recognition. The method includes receiving image data from an image; performing a multilayer feature extraction on the image data; generating current feature maps from the multilayer feature extraction; generating a region of proposal network map from at least the current feature maps, the region of proposal network map having one or more regions of interest; inputting previously generated feature maps and the current feature maps into a classifier with the region of proposed network map; and classifying the one or more regions of interest in the region of proposal network map.

Abstract: System and method for controlling a home assistant device include: receiving an audio input; performing speaker recognition on the audio input; in accordance with a determination that the audio input includes a voice input from a first user that is authorized to control the home assistant device: performing speech-to-text conversion on the audio input to obtain a textual string; and searching for a predefined trigger word for activating the home assistant device in the textual string; and in accordance with a determination that the audio input includes a voice input from the home assistant device: forgoing performance of speech-to-text conversion on the audio input; and forgoing search for the predefined trigger word.

Abstract: A method of performing automated object inspection includes obtaining a plurality of test images. For each of the plurality of test images, the method includes quantitatively determining a respective transformation from a predefined contour of the reference marker in the predefined common layout to a respective anchor contour corresponding to the reference marker as captured in the test image; and applying an inverse of the respective transformation to at least a portion of the test image to obtain a respective regularized version of the test image, such that the reference markers captured in the respective regularized versions of the plurality of test images share an image-independent location and orientation. The method further includes performing independent automated inspection on each of two or more sub-portions of the respective regularized versions of the test images.

Abstract: A method of performing automated object inspection includes obtaining a plurality of test images. For each of the plurality of test images, the method includes performing independent object inspection on each of two or more sub-portions of the test image. The method further includes segmenting the test image into at least a first sub-portion of the test image and a second sub-portion of the test image; performing object inspection on the first sub-portion of the test image using a first subset of information channels of the test image and a first model trained on a first set of training images containing the first component; and performing object inspection on the second sub-portion of the test image using a second subset of information channels of the test image, and a second model trained on a second set of training images containing the second component.

Abstract: A method, a non-transitory computer readable medium, and a system are disclosed for observing one or more subjects. The method includes monitoring a space with at least one master sensor, wherein a plurality of secondary sensors are installed in the space, and wherein a number of the at least one master sensor is less than a number of the plurality of secondary sensors; detecting regions of interest based on input from the at least one master sensor; identifying one or more secondary sensors from the plurality of secondary sensors in the detected regions of interest; and recognizing activities in the detected regions of interest from the one or more secondary sensors.

Abstract: A method of quality inspection is performed by a robotic arm that includes a plurality of segments, a camera at an end of the robotic arm, and a plurality of joints connecting two segments of the plurality of segments. The method includes (i) inspecting, via the camera, a surface of a product with the camera positioned at a first position, (ii) based on the inspecting, identifying: (a) an area of interest on the surface of the product, and (b) a relative location of the area of interest on the surface, (iii) positioning, based on the relative location of the area of interest on the surface, the camera at a second position, and (iv) inspecting, via the camera, the area of interest on the surface of the product with the camera positioned at the second position. Inspecting the area of interest includes inspecting a subset of the surface of the product.

Abstract: The disclosed computer-implemented method for artificial-intelligence-based automated surface inspection can include receiving customer data, a request for a targeted model, and compensation for the requested targeted model. The compensation can include an agreement to contribute the customer data and/or targeted model to be available for other third-party entities. The method can also include retrieving the pre-trained model from a pre-trained model pool. The pre-trained model can be related to objects in a second industry. The method can include generating the targeted model from the pre-trained model and the customer data. The targeted model can be related to mapping sensor data to surface anomalies. The method can also include providing the targeted model to the third-party entity. The method can further include updating a distributed blockchain structure to include the at least one of the customer data and the targeted model.

Abstract: The disclosed computer-implemented system and method for artificial-intelligence-based automated object identification and manipulation can include receiving a subsystem request from a third-party entity, the request being related to a subsystem for an object identification and manipulation system. The system and method also includes creating a developer request for a model suitable for the subsystem, the developer request including at least one approval condition. The system and method further includes evaluating a developer proposal received in response to the developer request, the developer proposal including a trained model, the evaluating including determining an accuracy level of the trained model, and the evaluating includes designating the trained model as an approved model if the developer proposal is approved. Also, the system and method includes providing the approved model to the third-party entity in response to the subsystem request.

Abstract: A voice command module is used to execute voice commands in a residential environment that contains a plurality of home devices. The voice command module includes a speech recognition module and command logic. The speech recognition module receives utterances from users and converts the utterances to commands from a vocabulary of predetermined commands, which includes a customization command to define a new wake-up utterance as corresponding to a wake-up command. The command logic executes the commands. When the customization command is received, the command logic changes the voice command module so that it now executes the wake-up command upon detection of the new wake-up utterance.

Abstract: A method for generating context-free grammar for intent deduction includes: retrieving, through a compiler, a first template file for a respective actionable intent corresponding to a first task that is to be performed by a machine; parsing the first template file, using the compiler, to identify a first nonterminal variable that refers to a first nonterminal file containing a plurality of first named entity values; and generating, based on a plurality of production rules specified in the first template file, including at least a first production rule that includes the first nonterminal variable, first context-free grammar for the respective actionable intent corresponding to the first task, wherein generating the first context-free grammar for the respective actionable intent corresponding to the first task includes rewriting the first production rule by replacing the first nonterminal variable in the first production rule with the plurality of first named entity values.

Abstract: A method of performing automated object inspection includes obtaining a plurality of test images. For each of the plurality of test images, the method includes performing independent object inspection on each of two or more sub-portions of the test image. The method further includes segmenting the test image into at least a first sub-portion of the test image and a second sub-portion of the test image; performing object inspection on the first sub-portion of the test image using a first subset of information channels of the test image and a first model trained on a first set of training images containing the first component; and performing object inspection on the second sub-portion of the test image using a second subset of information channels of the test image, and a second model trained on a second set of training images containing the second component.

Abstract: A method of performing automated object inspection includes obtaining a plurality of test images. For each of the plurality of test images, the method includes quantitatively determining a respective transformation from a predefined contour of the reference marker in the predefined common layout to a respective anchor contour corresponding to the reference marker as captured in the test image; and applying an inverse of the respective transformation to at least a portion of the test image to obtain a respective regularized version of the test image, such that the reference markers captured in the respective regularized versions of the plurality of test images share an image-independent location and orientation. The method further includes performing independent automated inspection on each of two or more sub-portions of the respective regularized versions of the test images.

Abstract: A method, a system, and a computer readable recording medium are disclosed for performing object recognition. The method includes receiving image data from an image; performing a multilayer feature extraction on the image data; generating current feature maps from the multilayer feature extraction; generating a region of proposal network map from at least the current feature maps, the region of proposal network map having one or more regions of interest; inputting previously generated feature maps and the current feature maps into a classifier with the region of proposed network map; and classifying the one or more regions of interest in the region of proposal network map.

Abstract: Systems and methods are provided for fall prediction. Such system may comprise one or more sensors configured to obtain at least gait information of a person, and one or more processors coupled to the one or more sensors. The one or more processors may be configured to receive the gait information of the person, obtain gait dynamic features based at least in part on the obtained gait information, and apply the obtained gait dynamic features to a learning model to predict at least a fall movement of the person.

Abstract: A massage robot uses machine vision to locate treatment spots for massage. In one approach, a massage robot includes one or more robotic arms, an image sensor and a control system. The image sensor captures images of the user. The control system includes an image processing module and a motion controller. The image processing module processes the images to locate a treatment spot on the user. The motion controller controls the robotic arm to perform a massage procedure on the identified treatment spot.

Abstract: A method, a computer readable recording medium, and a system are disclosed for image segmentation using controlled feedback in a neural network. The method includes extracting image data from an image; performing one or more semantic segmentations on the extracted image data; introducing one or more classifiers to each of the one or more semantic segmentations, each of the one or more classifiers assigning a probability to one or more classes of objects within the image; and generating a segmentation mask from the one or more semantic segmentations.

Abstract: A method for generating context-free grammar for intent deduction includes: retrieving, through a compiler, a first template file for a respective actionable intent corresponding to a first task that is to be performed by a machine; parsing the first template file, using the compiler, to identify a first nonterminal variable that refers to a first nonterminal file containing a plurality of first named entity values; and generating, based on a plurality of production rules specified in the first template file, including at least a first production rule that includes the first nonterminal variable, first context-free grammar for the respective actionable intent corresponding to the first task, wherein generating the first context-free grammar for the respective actionable intent corresponding to the first task includes rewriting the first production rule by replacing the first nonterminal variable in the first production rule with the plurality of first named entity values.