Abstract: Utilizing at least one existing policy (e.g. a manually engineered policy) for a robotic task, in generating reinforcement learning (RL) data that can be used in training an RL policy for an instance of RL of the robotic task. The existing policy can be one that, standing alone, will not generate data that is compatible with the instance of RL for the robotic task. In contrast, the generated RL data is compatible with RL for the robotic task at least by virtue of it including state data that is in a state space of the RL for the robotic task, and including actions that are in the action space of the RL for the robotic task. The generated RL data can be used in at least some of the initial training for the RL policy using reinforcement learning.

Abstract: A mobile robotic device includes a mobile base and a mast fixed relative to the mobile base. The mast includes a carved-out portion. The mobile robotic device further includes a three-dimensional (3D) lidar sensor mounted in the carved-out portion of the mast and fixed relative to the mast such that a vertical field of view of the 3D lidar sensor is angled downward toward an are in front of the mobile robotic device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using a neural network to predict locations of feeders in an electrical power grid. One of the methods includes training a generative adversarial network comprising a generator and a discriminator; and generating, by the generator, from input images, output images with feeder metadata that represents predicted locations of feeder assets, including receiving by the generator a first input image and generating by the generator a corresponding first output image with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

Abstract: Generating and utilizing action image(s) that represent a candidate pose (e.g., a candidate end effector pose), in determining whether to utilize the candidate pose in performance of a robotic task. The action image(s) and corresponding current image(s) can be processed, using a trained critic network, to generate a value that indicates a probability of success of the robotic task if component(s) of the robot are traversed to the particular pose. When the value satisfies one or more conditions (e.g., satisfies a threshold), the robot can be controlled to cause the component(s) to traverse to the particular pose in performing the robotic task.

Abstract: A method for matching production of FBA metabolism to supply and demand within a larger production network is described herein. An objective function of FBA metabolism is modified to include an upstream supply generated in upstream sub-units, as well as a downstream demand generated within downstream sub-units in the production network. FBA metabolism and the upstream and downstream sub-units are iteratively solved with updated initial conditions, producing a time series solution to the production network.

Abstract: A latent space is defined to represent sequences using training data and a machine-learning model. The training data identifies sequences of molecules and binding-approximation metrics that characterizes whether the molecules bind to a particular target and/or that approximate an extent to which the molecule is more likely to bind to the particular target than some other molecules. Supplemental training data is accessed that identifies other sequences of other molecules and binding affinity scores quantifying binding strengths between the molecules and the particular target. Projections of representations of the other sequences in the supplemental training data are projected in the latent space using the binding affinity scores. An area or position of interest within the latent space is identified based on the projections. A particular sequence represented within or at the area or position of interest or at the position of interest is identified for downstream processing.

Abstract: A system and method for automated lens adjustment for hyperspectral imaging is described. The system includes an image sensor and an electrically-controllable element arranged to set a spectral band for image capture by (i) selectively providing light for a selected spectral band or (ii) selectively filtering light to a selected spectral band. The system includes a tunable lens that is adjustable to change a focal length of the lens; and one or more data storage devices storing data that indicates different focus adjustment parameters corresponding to different spectral bands. The system includes a control system configured to perform operations including: selecting a spectral band; controlling the electrically-controllable element to set the spectral band for image capture; retrieving the focus adjustment parameter that corresponds to the spectral band; adjusting the lens based on the retrieved focus adjustment parameter; and capturing an image of the subject while the lens remains adjusted.

Abstract: Implementations are described herein for automatically generating quasi-realistic synthetic training images that are usable as training data for training machine learning models to perceive various types of plant traits in digital images. In various implementations, multiple labeled simulated images may be generated, each depicting simulated and labeled instance(s) of a plant having a targeted plant trait. In some implementations, the generating may include stochastically selecting features of the simulated instances of plants from a collection of plant assets associated with the targeted plant trait. The collection of plant assets may be obtained from ground truth digital image(s). In some implementations, the ground truth digital image(s) may depict real-life instances of plants having the target plant trait.

Abstract: An alert system is configured to track conditions for a tracking system. The one or more processors of the alert system are configured to receive data in a payload related to characteristics of tracking devices of the tracking system. The one or more processors are then configured to determine a condition timeframe for each condition for a given alert based on the data in the payload. The given alert requires at least a first condition and a second condition of the plurality of conditions. The one or more processors are configured to perform a query for historical data for the first condition based on a timestamp of the second condition that is prior to the first condition, determine an alert timeframe for the given alert based on the condition timeframes and the historical data, and store the given alert and the alert timeframe in a memory of the alert system.

Abstract: Systems and methods for designing a hybrid neural network comprising at least one physical neural network component and at least one digital neural network component. A loss function is defined within a design space composed of a plurality of voxels, the design space encompassing one or more physical structures of the at least one physical neural network component and one or more architectural features of the digital neural network. Values are determined for at least one functional parameter for the one or more physical structures, and the at least one architectural parameter for the one or more architectural features, using a domain solver to solve Maxwell's equations so that a loss determined according to the loss function is within a threshold loss. Final structures are defined for the at least one physical neural network component and the digital neural network component based on the values.

Abstract: The technology employs a state-based power control loop (PCL) architecture to maintain tracking and communication signal-to-noise ratios at suitable levels for optimal tracking performance and data throughput in a free-space optical communication system. Power for a link is adjustable to stay within a functional range of receiving sensors in order to provide continuous service to users. This avoids oversaturation and possible damage to the equipment. The approach can include decreasing or increasing the power to counteract a surge or drop while maintaining a near constant received power at a remote communication device. The system may receive power adjustment feedback from another communication terminal and perform state-based power control according to the received feedback. This can include re-initializing and reacquiring a link with the other communication terminal automatically after loss of power, without human intervention.

Abstract: A mobile robotic device is disclosed which includes a plurality of one-dimensional (1D) time-of-flight (ToF) sensors. Each 1D ToF sensor of the plurality of 1D ToF sensors may be mounted at a fixed position and orientation on the mobile robotic device. Each pair of 1D ToF sensors of the plurality of 1D ToF sensors may be fixed at respective positions and orientations relative to each other such that respective cones of coverage of the pair of 1D ToF sensors are non-overlapping.

Abstract: Methods and systems for using multiple hyperspectral cameras sensitive to different wavelengths to predict characteristics of objects for further processing, including recycling, are described. The multiple hyperspectral images can be used to predict higher resolution spectra by using a trained machine learning model. The higher resolution spectra may be more easily analyzed to sort plastics into a recyclability category. The hyperspectral images may also be used to identify and analyze dark or black plastics, which are challenging for SWIR, MWIR, and other wavelengths. The machine learning model may also predict the base polymers and contaminants of plastic objects for recycling. The hyperspectral images may be used to predict recyclability and other characteristics using a trained machine learning model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media for identification and re-identification of fish. In some implementations, first media representative of aquatic cargo is received. Second media based on the first media is generated, wherein a resolution of the second media is higher than a resolution of the first media. A cropped representation of the second media is generated. The cropped representation is provided to the machine learning model. In response to providing the cropped representation to the machine learning model, an embedding representing the cropped representation is generated using the machine learning model. The embedding is mapped to a high dimensional space. Data identifying the aquatic cargo is provided to a database, wherein the data identifying the aquatic cargo comprises an identifier of the aquatic cargo, the embedding, and a mapped region of the high dimensional space.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating an artificial neural network architecture based on a synaptic connectivity graph. According to one aspect, there is provided a method comprising: obtaining a synaptic resolution image of at least a portion of a brain of a biological organism; processing the image to identify: (i) a plurality of neurons in the brain, and (ii) a plurality of synaptic connections between pairs of neurons in the brain; generating data defining a graph representing synaptic connectivity between the neurons in the brain; determining an artificial neural network architecture corresponding to the graph representing the synaptic connectivity between the neurons in the brain; and processing a network input using an artificial neural network having the artificial neural network architecture to generate a network output.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for implementing a reservoir computing neural network. In one aspect there is provided a reservoir computing neural network comprising: (i) a brain emulation sub-network, and (ii) a prediction sub-network. The brain emulation sub-network is configured to process the network input in accordance with values of a plurality of brain emulation sub-network parameters to generate an alternative representation of the network input. The prediction sub-network is configured to process the alternative representation of the network input in accordance with values of a plurality of prediction sub-network parameters to generate the network output. The values of the brain emulation sub-network parameters are determined before the reservoir computing neural network is trained and are not adjusting during training of the reservoir computing neural network.

Abstract: Methods, systems, and apparatus for an infrared and visible imaging system. In some implementations, Image data from a visible-light camera is obtained. A position of a device is determined based at least in part on the image data from the visible-light camera. An infrared camera is positioned so that the device is in a field of view of the infrared camera, with the field of view of the infrared camera being narrower than the field of view of the visible-light camera. Infrared image data from the infrared camera that includes regions representing the device is obtained. Infrared image data from the infrared camera that represents the device is recorded. Position data is also recorded that indicates the location and pose of the infrared camera when the infrared image data is acquired by the infrared camera.

Abstract: This disclosure describes a system and method for generating a subsurface model representing lithological characteristics and attributes of the subsurface of a celestial body or planet. By automatically ingesting data from many sources, a machine learning system can infer information about the characteristics of regions of the subsurface and build a model representing the subsurface rock properties. In some cases, this can provide information about a region using inferred data, where no direct measurements have been taken. Remote sensing data, such as aerial or satellite imagery, gravimetric data, magnetic field data, electromagnetic data, and other information can be readily collected or is already available at scale.

Abstract: Techniques for improving a customer experience include presenting a selectable component to a user through a graphical user interface of a computing device that is communicably coupled to a server system through a network, the selectable component particularly associated with the user, the server system associated with a vendor and including a plurality of vendor product descriptions; receiving an activation of the selectable component from the user; based on the activation, identifying a plurality of profile data associated with the user; culling the plurality of vendor product descriptions to generate, based on the plurality of profile data, a subset of vendor product descriptions from the plurality of vendor product descriptions; and presenting the subset of vendor product descriptions to the user through the graphical user interface.

Abstract: An electroencephalography (EEG) headset can include an arrangement of straps that provides the ability to adjust the size and shape of the headset once disposed on a user's head. In some implementations, the headset can include a first elastic strap extending from a first side of the headset to a second side of the headset along a topside of the headset. The headset can also include a second strap including at least one inelastic portion and at least one elastic portion, at least one EEG electrode coupled to the second strap, a third elastic strap extending from the first side of the headset to the second side of the headset along an underside of the headset, and a plurality of connectors that couple the elastic first strap, the second strap, or the third elastic strap.