Abstract: A method includes receiving image data representing an environment of a robotic device from a camera on the robotic device. The method further includes applying a trained dense network to the image data to generate a set of feature values, where the trained dense network has been trained to accomplish a first robot vision task. The method additionally includes applying a trained task-specific head to the set of feature values to generate a task-specific output to accomplish a second robot vision task, where the trained task-specific head has been trained to accomplish the second robot vision task based on previously generated feature values from the trained dense network, where the second robot vision task is different from the first robot vision task. The method also includes controlling the robotic device to operate in the environment based on the task-specific output generated to accomplish the second robot vision task.

Abstract: A method for producing an experimental output satisfying an objective includes conducting an experimental execution process including applying a selection criterion to select an approach to determining a set of parameters for a set of experiments, and determining a first set of parameters for a first experiment according to the selected approach based on one or more of (i) a predicted relationship between a set of parameters and a characteristic of a corresponding experimental output, (ii) the measured characteristic of a second experimental output from a second experiment executed according to a second set of parameters, (iii) the objective, and (iv) a parameter selection rule.

Abstract: Methods, apparatus, systems, and computer-readable media are provided for visually annotating rendered multi-dimensional representations of robot environments. In various implementations, an entity may be identified that is present with a telepresence robot in an environment. A measure of potential interest of a user in the entity may be calculated based on a record of one or more interactions between the user and one or more computing devices. In some implementations, the one or more interactions may be for purposes other than directly operating the telepresence robot. In various implementations, a multi-dimensional representation of the environment may be rendered as part of a graphical user interface operable by the user to control the telepresence robot. In various implementations, a visual annotation may be selectively rendered within the multi-dimensional representation of the environment in association with the entity based on the measure of potential interest.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a storage device, for electric grid asset detection are enclosed. An electric grid asset detection method includes: obtaining overhead imagery of a geographic region that includes electric grid wires; identifying the electric grid wires within the overhead imagery; and generating a polyline graph of the identified electric grid wires. The method includes replacing curves in polylines within the polyline graph with a series of fixed lines and endpoints; identifying, based on characteristics of the fixed lines and endpoints, a location of a utility pole that supports the electric grid wires; detecting an electric grid asset from street level imagery at the location of the utility pole; and generating a representation of the electric grid asset for use in a model of the electric grid.

Abstract: A nuclear radiation detector is disclosed. The detector includes a housing including therein: a scintillator; and a multi-pixel optical sensor positioned, relative to the scintillator, to receive photons emitted by the scintillator in response to interactions with nuclear radiation. The housing isolates the scintillator and the multi-pixel optical sensor from external light. The detector includes one or more processors operably connectable to the multi-pixel optical sensor; and one or more data stores coupled to the processors having instructions stored thereon which cause the processors to perform operations.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, that validate the synchronization of controllers in an aquaculture environment. One of the methods includes an image processor that receives images generated by a first image generating device that includes a light filter that is associated with light of a particular light frequency while an aquaculture environment was illuminated with light. Based on the image, the image processor determines whether the intensity value of the light frequency in the image satisfies a threshold value. Based on determining whether the intensity value of the light frequency in the image satisfies the threshold value, the image processor determines whether the aquaculture environment was illuminated with light of the particular light frequency when the image was generated. The image processor provides an indication of whether the aquaculture was illuminated with light of the particular frequency when the image was generated.

Abstract: Methods, systems, and computer programs encoded on a computer storage medium, for improving EEG measurements by identifying artifacts present in EEG measurements and providing a real-time indication to a user of likely artifacts in EEG measurements are described. EEG measurements of a patient can be obtained by placing a wearable device or EEG cap on a patient's head. Sensors in the cap provide EEG data to a computing device that processes the data to identify one or more artifacts in the EEG data. The artifacts can be identified by conducting one or more operations of determining the signal to noise ratio of the line noise, calculating mutual information between sensor pairs, and applying the p-welch method. Based on the types of artifacts identified, the computing device can output an indicator that provides feedback to the technician performing an EEG test to make adjustments to the test setup.

Abstract: Techniques are described herein for robotic control using value distributions. In various implementations, as part of performing a robotic task, state data associated with the robot in an environment may be generated based at least in part on vision data captured by a vision component of the robot. A plurality of candidate actions may be sampled, e.g., from continuous action space. A trained critic neural network model that represents a learned value function may be used to process a plurality of state-action pairs to generate a corresponding plurality of value distributions. Each state-action pair may include the state data and one of the plurality of sampled candidate actions. The state-action pair corresponding to the value distribution that satisfies one or more criteria may be selected from the plurality of state-action pairs. The robot may then be controlled to implement the sampled candidate action of the selected state-action pair.

Abstract: In embodiments, acquiring sensor data associated with a plant growing in a field, and analyzing the sensor data to extract one or more phenotypic traits associated with the plant from the sensor data. Indexing the one or more phenotypic traits to one or both of an identifier of the plant or a virtual representation of a part of the plant, and determining one or more plant insights based on the one or more phenotypic traits, wherein the one or more plant insights includes information about one or more of a health, a yield, a planting, a growth, a harvest, a management, a performance, and a state of the plant. One or more of the health, yield, planting, growth, harvest, management, performance, and the state of the plant are included in a plant insights report that is generated.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining an artificial neural network architecture corresponding to a sub-graph of a synaptic connectivity graph. In one aspect, there is provided a method comprising: obtaining data defining a graph representing synaptic connectivity between neurons in a brain of a biological organism; determining, for each node in the graph, a respective set of one or more node features characterizing a structure of the graph relative to the node; identifying a sub-graph of the graph, comprising selecting a proper subset of the nodes in the graph for inclusion in the sub-graph based on the node features of the nodes in the graph; and determining an artificial neural network architecture corresponding to the sub-graph of the graph.

Abstract: Methods and apparatus related to receiving a request that includes robot instructions and/or environmental parameters, operating each of a plurality of robots based on the robot instructions and/or in an environment configured based on the environmental parameters, and storing data generated by the robots during the operating. In some implementations, at least part of the stored data that is generated by the robots is provided in response to the request and/or additional data that is generated based on the stored data is provided in response to the request.

Abstract: Image data is obtained that indicates an extent to which one or more objects reflect, scatter, or absorb light at each of multiple wavelength bands, where the image data was collected while a conveyor belt was moving the object(s). The image data is preprocessed by performing an analysis across frequencies and/or performing an analysis across a representation of a spatial dimension. A set of feature values is generated using the image preprocessed image data. A machine-learning model generates an output using to the feature values. A prediction of an identity of a chemical in the one or more objects or a level of one or more chemicals in the object(s) is generated using the output. Data is output indicating the prediction of the identity of the chemical in the object(s) or the level of the one or more chemicals in at least one of the one or more objects.

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: Implementations are described herein for analyzing a sequence of digital images captured by a mobile vision sensor (e.g., integral with a robot), in conjunction with information (e.g., ground truth) known about movement of the vision sensor, to determine spatial dimensions of object(s) and/or an area captured in a field of view of the mobile vision sensor. Techniques avoid the use of visual indicia of known dimensions and/or other conventional tools for determining spatial dimensions, such as checkerboards. Instead, techniques described herein allow spatial dimensions to be determined using less resources, and are more scalable than conventional techniques.

Abstract: Generating a robot control policy that regulates both motion control and interaction with an environment and/or includes a learned potential function and/or dissipative field. Some implementations relate to resampling temporally distributed data points to generate spatially distributed data points, and generating the control policy using the spatially distributed data points. Some implementations additionally or alternatively relate to automatically determining a potential gradient for data points, and generating the control policy using the automatically determined potential gradient. Some implementations additionally or alternatively relate to determining and assigning a prior weight to each of the data points of multiple groups, and generating the control policy using the weights.

Abstract: Methods, systems, and apparatus, including computer-readable storage devices, for robot navigation using 2D and 3D path planning. In the disclosed method, a robot accesses map data indicating two-dimensional layout of objects in a space and evaluates candidate paths for the robot to traverse. In response to determining that the candidate paths do not include a collision-free path across the space for a two-dimensional profile of the robot, the robot evaluates a three-dimensional shape of the robot with respect to a three-dimensional shape of an object in the space. Based on the evaluation of the three-dimensional shapes, the robot determines a collision-free path to traverse through the space.

Abstract: Implementations are described herein for analyzing vision data depicting undesirable plants such as weeds to detect various attribute(s). The detected attribute(s) of a particular undesirable plant may then be used to select, from a plurality of available candidate remediation techniques, the most suitable remediation technique to eradicate or otherwise eliminate the undesirable plants.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for causing a transducer to transmit an acoustic input signal into a member of a device. Receiving a detection signal representing reverberations of the input signal traveling within the member from a receiver. Detecting a contact of the member with an object external to the member based on a change in the detection signal, where the change in the detection signal represents an alteration in the reverberations of the input signal caused by the contact of the member with the object. Determining a position along the member of a point of the contact of the member with the object based on the change in the detection signal.

Abstract: At least one machine-accessible storage medium that provides instructions that, when executed by a machine, will cause the machine to perform operations. The operations comprise configuring a simulated environment to be representative of a physical device based, at least in part, on an initial description of the physical device that described structural parameters of the physical device. The operations further comprise performing a physics simulation with an artificial intelligence (“AI”) accelerator. The AI accelerator includes a matrix multiply unit for computing convolution operations via a plurality of multiply-accumulate units. The operations further comprise computing a field response in response of the physical device in response to an excitation source within the simulated environment when performing the physics simulation. The field response is computed, at least in part, with the convolution operations to perform spatial differencing.

Abstract: Implementations are described herein for automatically generating synthetic training images that are usable as training data for training machine learning models to detect, segment, and/or classify various types of plants in digital images. In various implementations, a digital image may be obtained that captures an area. The digital image may depict the area under a lighting condition that existed in the area when a camera captured the digital image. Based at least in part on an agricultural history of the area, a plurality of three-dimensional synthetic plants may be generated. The synthetic training image may then be generated to depict the plurality of three-dimensional synthetic plants in the area. In some implementations, the generating may include graphically incorporating the plurality of three-dimensional synthetic plants with the digital image based on the lighting condition.