Abstract: In one embodiment, a method includes receiving, from a first sensor on a robot, first sensor data indicative of an environment of the robot. The method also includes identifying, based on the first sensor data, an object of an object type in the environment of the robot, where the object type is associated with a classifier that takes sensor data from a predetermined pose relative to the object as input. The method further includes causing the robot to position a second sensor on the robot at the predetermined pose relative to the object. The method additionally includes receiving, from the second sensor, second sensor data indicative of the object while the second sensor is positioned at the predetermined pose relative to the object. The method further includes determining, by inputting the second sensor data into the classifier, a property of the object.

Abstract: Methods, systems, and apparatus, including computer-readable media storing executable instructions, for enhancing robot learning. In some implementations, a robot stores first embeddings generated using a first machine learning model, and the first embeddings include one or more first private embeddings that are not shared with other robots. The robot receives a second machine learning model from a server system over a communication network. The robot generates a second private embedding for each of the one or more first private embeddings using the second machine learning model. The robot adds the second private embeddings to the cache of the robot and removes the one or more first private embeddings from the cache of the robot.

Abstract: An apparatus includes a holder to support an indenter relative to a sample, a depth sensor, and a controller. The operations include applying a first force on the sample with the indenter and determining a first depth of the indenter based on data generated by the sensor, moving the indenter from the first depth to a greater predetermined depth, then applying the first force on the sample with the indenter and determining a second depth of the indenter based on second data generated by the sensor, and determining a value indicative of hardness of the sample based on a difference between the first depth and the second depth. The apparatuses described can use a single scale for hardness that enables hardness values for different materials to be compared to one another.

Abstract: Methods, systems, and apparatus for receiving a command for controlling a robot, the command referencing an object, receiving sensor data for a portion of an environment of the robot, identifying, from the sensor data, a gesture of a human that indicates a spatial region located outside of the portion of the environment described by the sensor data, searching map data for the object, determining, based at least on searching the map data for the object referenced in the command, that the object referenced in the command is present in the spatial region, and in response to determining that the object referenced in the command is present in the spatial region, controlling the robot to perform an action with respect to the object referenced in the command.

Abstract: Methods and systems for modifying the inertial parameters used in a virtual robot model that simulates the interactions of a real-world robot with an environment to better reflect the actual inertial properties of the real-world robot. In one aspect, a method includes obtaining joint physical parameter measurements for the joints of a real-world robot, determining simulated joint physical parameter values for each of the joint physical parameter measurements, and adjusting an estimate of inertial properties of the real-world robot used by the virtual robot dynamic model to reduce a difference between the simulated joint physical parameter values and the corresponding joint physical parameter measurements.

Abstract: Example methods and systems for determining 3D scene geometry by projecting patterns of light onto a scene are provided. In an example method, a first projector may project a first random texture pattern having a first wavelength and a second projector may project a second random texture pattern having a second wavelength. A computing device may receive sensor data that is indicative of an environment as perceived from a first viewpoint of a first optical sensor and a second viewpoint of a second optical sensor. Based on the received sensor data, the computing device may determine corresponding features between sensor data associated with the first viewpoint and sensor data associated with the second viewpoint. And based on the determined corresponding features, the computing device may determine an output including a virtual representation of the environment that includes depth measurements indicative of distances to at least one object.

Abstract: Example implementations may relate to providing a dynamic jig in a three-dimensional (3D) coordinate system. Specifically, a control system may (i) receive task data specifying a manipulation of one or more parts at a specified location; (ii) determine: (a) one or more work surfaces and (b) a first position of each of the one or more work surfaces, such that the one or more work surfaces collectively provide a jig to facilitate the specified manipulation of the parts; (iii) a plurality of guide end effectors that are positionable by one or more robotic devices such that the end effectors provide the work surfaces at the respectively determined first positions; and (iv) operate the one or more robotic devices to position the guide end effectors to provide the one or more work surfaces at the respectively determined first positions, thereby forming the jig from the one or more work surfaces.

Abstract: Techniques for generating solutions from aural inputs include identifying, with one or more machine learning engines, a plurality of aural signals provided by two or more human speakers, at least some of the plurality of aural signals associated with a human-perceived problem; parsing, with the one or more machine learning engines, the plurality of aural signals to generate a plurality of terms, each of the terms associated with the human-perceived problem; deriving, with the one or more machine learning engines, a plurality of solution sentiments and a plurality of solution constraints from the plurality of terms; generating, with the one or more machine learning engines, at least one solution to the human-perceived problem based on the derived solution sentiments and solution constraints; and presenting the at least one solution of the human-perceived problem to the two or more human speakers through at least one of a graphical interface or an auditory interface.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, of training a global policy neural network. One of the methods includes initializing a plurality of instances of the robotic task. For each instance of the robotic task, the method includes generating a trajectory of state-action pairs by selecting actions to be performed by the robotic agent while performing the instance of the robotic task in accordance with current values of the parameters of the global policy neural network, and optimizing a local policy controller that is specific to the instance on the trajectory of state-action pairs for the instance. The method further includes generating training data for the global policy neural network using the local policy controllers, and training the global policy neural network on the training data to adjust the current values of the parameters of the global policy neural network.

Abstract: Methods and systems for detecting sensor orientation characteristics using marker-based localization are disclosed herein. In one aspect, a robotic device can: receive a map of a horizontal marker plane that includes mapped positions of a first marker and a second marker arranged in the horizontal marker plane; receive, from a sensor configured to scan a two-dimensional sensor plane, sensor data indicative of positions of the first and second markers relative to the sensor; determine measured positions of the first and second markers based on the sensor data and a current position of the sensor; determine a difference vector between a first vector that connects the mapped positions of the first and second markers and a second vector that connects the measured positions of the first and second markers; and determine, based on the difference vector, an orientation of the two-dimensional sensor plane relative to the horizontal marker plane.

Abstract: A bioamplifier for analyzing electroencephalogram (EEG) signals is disclosed. The bioamplifier includes an input terminal for receiving an EEG signal from a plurality of sensors coupled to a user. The bioamplifier also includes an analogue-to-digital converter arranged to receive the EEG signal from the input terminal and convert the EEG signal to a digital EEG signal. A data processing apparatus within the bioamplifier is arranged to receive the digital EEG signal from the analogue-to-digital converter and programmed to process, in real time the digital EEG signal using a first machine learning model to generate a cleaned EEG signal having a higher signal-to-noise ratio than the digital EEG signal. The bioamplifier further includes a power source to provide electrical power to the analogue-to-digital converter and the data processing apparatus. The bioamplifier includes a housing that contains the analogue-to-digital converter, the data processing apparatus, the power source, and the sensor input.

Abstract: In some embodiments, an ear-mounted sound reproduction system is provided. The system includes an ear-mountable housing that sits within the pinna of the ear and occludes the ear canal. In some embodiments, the ear-mountable housing includes a plurality of external-facing microphones. Because the external-facing microphones may be situated within the pinna of the ear but outside of the ear canal, the microphones will experience some, but not all, of the three-dimensional acoustic effects of the pinna. In some embodiments, sound is reproduced by an internal-facing driver element of the housing using a plurality of filters applied to the signals received by the plurality of external-facing microphones to preserve three-dimensional localization cues that would be present at the eardrum in the absence of the housing, such that the housing is essentially transparent to the user. In some embodiments, techniques are provided for deriving the plurality of filters.

Abstract: Techniques described herein relate to using reduced-dimensionality embeddings generated from robot sensor data to identify predetermined semantic labels that guide robot interaction with objects. In various implementations, obtaining, from one or more sensors of a robot, sensor data that includes data indicative of an object observed in an environment in which the robot operates. The sensor data may be processed utilizing a first trained machine learning model to generate a first embedded feature vector that maps the data indicative of the object to an embedding space. Nearest neighbor(s) of the first embedded feature vector may be identified in the embedding space. Semantic label(s) may be identified based on the nearest neighbor(s). A given grasp option may be selected from enumerated grasp options previously associated with the semantic label(s). The robot may be operated to interact with the object based on the pose and using the given grasp option.

Abstract: Systems and methods are provided for specifying safety rules for robotic devices. A computing device can determine information about any actors present within a predetermined area of an environment. The computing device can determine a safety classification for the predetermined area based on the information. The safety classification can include: a low safety classification if the information indicates zero actors are present within the predetermined area, a medium safety classification if the information indicates any actors are present within the predetermined area all are of a predetermined first type, and a high safety classification if the information indicates at least one actor present within the predetermined area is of a predetermined second type. After determining the safety classification for the predetermined area, the computing device can provide a safety rule for operating within the predetermined area to a robotic device operating in the environment.

Abstract: Implementations are described herein for offline computation and caching of precalculated joint trajectories. In various implementations, an instruction may be obtained to move an end effector of a robot between start and target positions. A first type of trajectory planning may be performed in real time or “online” to calculate a first joint trajectory of the robot that moves the end effector from the start to target position. The robot may then implement the first joint trajectory. A second type of trajectory planning may be performed offline, e.g., during downtime of the robot, to precalculate a second joint trajectory of the robot to move the end effector from the start to target position. The second type of trajectory planning may require more resources than were required by the first type of trajectory planning. Data indicative of the precalculated second joint trajectory of the robot may be stored for future use.

Abstract: Implementations relate to diagnosis of crop yield predictions and/or crop yields at the field- and pixel-level. In various implementations, a first temporal sequence of high-elevation digital images may be obtained that captures a geographic area over a given time interval through a crop cycle of a first type of crop. Ground truth operational data generated through the given time interval and that influences a final crop yield of the first geographic area after the crop cycle may also be obtained. Based on these data, a ground truth-based crop yield prediction may be generated for the first geographic area at the crop cycle's end. Recommended operational change(s) may be identified based on distinct hypothetical crop yield prediction(s) for the first geographic area. Each distinct hypothetical crop yield prediction may be generated based on hypothetical operational data that includes altered data point(s) of the ground truth operational data.

Abstract: A photovoltaic power system includes a photofuel having a molecular structure to emit light, and a receptacle including the photofuel disposed within. One or more photovoltaic cells are positioned within the receptacle to receive light emitted from the photofuel, and a negative electrode is coupled to the one or more photovoltaic cells. A positive electrode is coupled to the one or more photovoltaic cells to produce an electrical potential between the negative electrode and the positive electrode when a photocurrent is generated by the one or more photovoltaic cells in response to the one or more photovoltaic cells receiving the light emitted from the photofuel.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving, by one or more non-real-time processors, data defining a light illumination pattern for a robotic device. Generating, by the one or more non-real-time processors and based on the data, a spline that represents the light illumination pattern, where a knot vector of the spline defines a timing profile of the light illumination pattern. Providing the spline to one or more real-time processors of the robotic system. Calculating, by the one or more real-time processors, an illumination value from the spline at each of a plurality of time steps. Controlling, by the one or more real-time processors, illumination of a lighting display of the robotic system in accordance with the illumination value of the spline at each respective time step.