Abstract: Aspects of the disclosure provide for identifying problematic areas within a service area for an autonomous vehicle transportation service. For instance, a starting location within the service area corresponding to a potential pickup location for passengers or cargo for the service may be identified. A destination within the service area may be identified. A simulation may be run in order to determine a route for a simulated vehicle to travel between the starting location and the destination. That the route includes a particular type of maneuver may be determined. A new simulation without allowing the simulated vehicle to complete the particular type of maneuver may be run. Whether the simulated vehicle reaches the destination in the new simulation may be determined. Based on the determination of whether the simulated vehicle reaches the destination in the new simulation, the starting location and destination location may be flagged as potentially problematic areas.

Abstract: The technology relates to localizing a vehicle. As one approach, a first LIDAR sensor scan data of an environment of the vehicle and localization data for the vehicle at a location where the first LIDAR sensor scan was captured are stored. Thereafter, the computing device is suspended and subsequently unsuspended. After the computing device is unsuspended, second LIDAR sensor scan data of the vehicle's environment is received. The first LIDAR sensor scan data is compared to the second LIDAR sensor scan data to determine whether the vehicle has moved. Based on the determination of whether the vehicle has moved from the location, the stored localization data is used to localize the vehicle. Other approaches are also described.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining a location of a particular object relative to a vehicle. In one aspect, a method includes obtaining sensor data captured by one or more sensors of a vehicle. The sensor data is processed by a convolutional neural network to generate a sensor feature representation of the sensor data. Data is obtained which defines a particular spatial region in the sensor data that has been classified as including sensor data that characterizes the particular object. An object feature representation of the particular object is generated from a portion of the sensor feature representation corresponding to the particular spatial region. The object feature representation of the particular object is processed using a localization neural network to generate the location of the particular object relative to the vehicle.

Abstract: This technology relates to a system for clearing sensors. The system may include an air knife configured to clear a sensor housing of debris. The air knife may be attached to a bearing comprising a bearing ring. The system may also include a motor, wherein the motor is configured to rotate the bearing ring around the sensor housing.

Abstract: A Printed Circuit Board (PCB) includes a via extending through at least one layer of the PCB. The PCB may also include a first catch pad connected to the via and located within a first metal layer of the PCB. The first catch pad may have a first size. The PCB may further include a second catch pad connected to the via and located within a second metal layer of the PCB. The second catch pad may have a second size greater than the first size. The second catch pad may overlap horizontally with a portion of a metallic feature in the first metal layer to obstruct light incident on a first side of the PCB from transmission to a second side of the PCB through a region of dielectric material near the via.

Abstract: In some implementations, an image classification system of an autonomous or semi-autonomous vehicle is capable of improving multi-object classification by reducing repeated incorrect classification of objects that are considered rarely occurring objects. The system can include a common instance classifier that is trained to identify and recognize general objects (e.g., commonly occurring objects and rarely occurring objects) as belonging to specified object categories, and a rare instance classifier that is trained to compute one or more rarity scores representing likelihoods that an input image is correctly classified by the common instance classifier. The output of the rare instance classifier can be used to adjust the classification output of the common instance classifier such that the likelihood of input images being incorrectly classified is reduced.

Abstract: Disclosed herein are neural networks for generating target classifications for an object from a set of input sequences. Each input sequence includes a respective input at each of multiple time steps, and each input sequence corresponds to a different sensing subsystem of multiple sensing subsystems. For each time step in the multiple time steps and for each input sequence in the set of input sequences, a respective feature representation is generated for the input sequence by processing the respective input from the input sequence at the time step using a respective encoder recurrent neural network (RNN) subsystem for the sensing subsystem that corresponds to the input sequence. For each time step in at least a subset of the multiple time steps, the respective feature representations are processed using a classification neural network subsystem to select a respective target classification for the object at the time step.

Abstract: The present disclosure relates to optical systems and methods for their manufacture. An example optical system includes a first substrate having a mounting surface and a spacer structure having at least one cavity. The spacer structure is coupled to the mounting surface of the first substrate. The optical system also includes a light-emitter device that is coupled to the spacer structure and a detector device coupled to the first substrate such that the at least one detector device is disposed within the at least one cavity of the spacer structure. The optical system also includes a second substrate that mounts a lens and a waveguide and is coupled to the spacer structure. The optical system also includes a shim coupled between the second surface of the spacer structure and a mounting surface of the second substrate.

Abstract: An example method includes using a plurality of switches corresponding to a plurality of capacitors to select a first set of capacitors for charging at a first time. Charging the first set of capacitors corresponds to sampling from a first set of adjacent light detectors. The method includes using the plurality of switches to select a second set of capacitors from the plurality of capacitors for discharging at a second time. The method includes using a sampling switch to sample an output of the second set of capacitors as they discharge. The output of the second set of capacitors corresponds to the first set of adjacent light detectors. The method includes determining, based on sampling the output of the second set of capacitors, a collective intensity of light received by the first set of adjacent light detectors.

Abstract: Methods, systems, and techniques for validating road intersections are presented. The method includes receiving annotated intersection data that, for each lane of a particular candidate intersection, indicates one or more attributes of the lane; processing the annotated intersection data to generate an intersection fingerprint for the particular candidate intersection that is sensitive to a topological structure of the lanes of the particular candidate intersection and is invariant to a geometry of the lanes of the particular candidate intersection; designating the particular candidate intersection as validated without performing the one or more quality control processes based at least on the intersection fingerprint; and storing a representation of the particular candidate intersection in the roadgraph.

Abstract: Methods and systems for use of a reference image to detect a road obstacle are described. A computing device configured to control a vehicle, may be configured to receive, from an image-capture device, an image of a road on which the vehicle is travelling. The computing device may be configured to compare the image to a reference image; and identify a difference between the image and the reference image. Further, the computing device may be configured to determine a level of confidence for identification of the difference. Based on the difference and the level of confidence, the computing device may be configured to modify a control strategy associated with a driving behavior of the vehicle; and control the vehicle based on the modified control strategy.

Abstract: Aspects of the disclosure provide for determining when to provide and providing secondary disengage alerts for a vehicle having autonomous and manual driving modes. For instance, while the vehicle is being controlled in the autonomous driving mode, receiving user input at one or more using input devices of the vehicle. In response to receiving the user input, the vehicle may be transitioned from the autonomous driving mode to a manual driving mode and provide a primary disengage alert to an occupant of the vehicle regarding the transition. Whether to provide a secondary disengage alert may be determined based on at least circumstances of the user input. After the transition, the secondary disengage alert may be provided based on the determination.

Abstract: A system and method for avoiding or mitigating collisions by an autonomous vehicle is provided. Raw sensor data for the vehicle's environment is received, such as through a perception system of the vehicle. A trajectory is also received, the trajectory having a width corresponding to a distance that is at least as wide as the vehicle. The trajectory and raw sensor data are projected onto a grid, such as a static occupancy grid, having a plurality of cells. It is determined whether any cell that overlaps with the trajectory includes the raw sensor data. It is further determined, based on the raw sensor data, whether an obstacle is present along the trajectory. When no obstacles are determined to be present along the trajectory, the trajectory may be verified. However, where obstacles are determined to be present, the vehicle may take an action to avoid collision.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for implementing an interactive autonomous vehicle agent. One of the methods includes receiving a request to generate an experience tuple for a vehicle in a particular driving context. A predicted environment observation representing a predicted environment of the autonomous vehicle after the candidate action is taken by the autonomous vehicle in an initial environment is generated, including providing an initial environment observation and the candidate action as input to a vehicle behavior model neural network trained to generate predicted environment observations. An immediate quality value is generated from a context-specific quality model that generates immediate quality values that are specific to the particular driving context.

Abstract: A vehicle is provided that may distinguish between dynamic obstacles and static obstacles. Given a detector for a class of static obstacles or objects, the vehicle may receive sensor data indicative of an environment of the vehicle. When a possible object is detected in a single frame, a location of the object and a time of observation of the object may be compared to previous observations. Based on the object being observed a threshold number of times, in substantially the same location, and within some window of time, the vehicle may accurately detect the presence of the object and reduce any false detections.

Abstract: A radar system in an autonomous vehicle may be operated in various modes and with various configurations. In one example, the radar system determines a target range for further interrogation. The target range may be determined based on the radar system transmitting a first electromagnetic radiation signal and receiving a first reflected electromagnetic signal radiation signal. After the radar system determines a target range, it transmits a second electromagnetic radiation signal. Additionally, the radar system receives a reflected electromagnetic signal radiation based on the transmission. After receiving the reflected signal, the radar system can process the reflected signal to only have components associated with the target range. The processing of the reflected signal may create a processed signal. Finally, the radar system may determine at least one parameter of a target object based on the processed signal.

Abstract: A LIDAR device includes an input node, an output node, and a sample-and-convert circuit. The input node receives a photodetector signal, and the output node generates an output signal indicating a light intensity value of the photodetector signal. The sample-and-convert circuit includes a number of detection channels coupled in parallel between the input node and the output node. In some aspects, each of the detection channels may be configured to sample a value of the photodetector signal during the sample mode and to hold the sampled value during the convert mode using a single capacitor.