Abstract: Aspects and implementations of the present disclosure address shortcomings of the existing technology by enabling motion pattern-assisted object classification of objects in an environment of an autonomous vehicle (AV) by obtaining, from a sensing system of the AV, a plurality of return points, each return point comprising one or more velocity values and one or more coordinates of a reflecting region that reflects a signal emitted by the sensing system, identifying an association of the plurality of return points with an object in an environment of the AV, identifying, in view of the one or more velocity values of at least some of the plurality of return points, a type of the object or a type of a motion of the object, and causing a driving path of the AV to be determined in view of the identified type of the object.

Abstract: Example embodiments relate to self-reflection filtering techniques within radar data. A computing device may use radar data to determine a first radar representation that conveys information about surfaces in a vehicle's environment. The computing device may use a predefined model to generate a second radar representation that assigns predicted self-reflection values to respective locations of the environment based on the information about the surfaces conveyed by the first radar representation. The predefined model can enable a predefined self-reflection value to be assigned to a first location based on information about a surface positioned at a second location and a relationship between the first location and the second location. The computing device may then modify the first radar representation based on the predicted self-reflection values in the second radar representation and provide instructions to a control system of the vehicle based on modifying the first radar representation.

Abstract: Aspects and implementations are related to systems and techniques enabling predictions of a motion change in a moving vehicle, predictions of an onset of a motion of an idling vehicle, and classification of vehicles based, at least in part, on vibrometry data obtained using light detection and ranging devices.

Abstract: The technology relates to developing a highly accurate understanding of a vehicle's sensor fields of view in relation to the vehicle itself. A training phase is employed to gather sensor data in various situations and scenarios, and a modeling phase takes such information and identifies self-returns and other signals that should either be excluded from analysis during real-time driving or accounted for to avoid false positives. The result is a sensor field of view model for a particular vehicle, which can be extended to other similar makes and models of that vehicle. This approach enables a vehicle to determine when sensor data is of the vehicle or something else. As a result, the detailed modeling allowing the on-board computing system to make driving decisions and take other actions based on accurate sensor information.

Abstract: Example embodiments relate to radar image video compression techniques using per-pixel Doppler measurements, which can involve initially receiving radar data from a radar unit to generate a radar representation that represents surfaces in the environment. Based on Doppler scores in the radar representation, a range rate can be determined for each pixel that indicates a radial direction motion for a surface represented by the pixel. The range rates and backscatter values can then be used to estimate a radar representation prediction for subsequent radar data received from the radar unit, which enables a generation of a compressed radar data file that represents the difference between the radar representation prediction and the actual representation determined for the subsequent radar data. The compressed radar data file can be stored in memory, transmitted to other devices, and decompressed and used to train models via machine learning.

Abstract: One example method involves rotating a housing of a light detection and ranging (LIDAR) device about a first axis. The housing includes a first optical window and a second optical window. The method also involves transmitting a first plurality of light pulses through the first optical window to obtain a first scan of a field-of-view (FOV) of the LIDAR device. The method also involves transmitting a second plurality of light pulses through the second optical window to obtain a second scan of the FOV. The method also involves identifying, based on the first scan and the second scan, a portion of the FOV that is at least partially occluded by an occlusion.

Abstract: Aspects of the disclosure provide for automatically generating labels for sensor data. For instance first sensor data for a first vehicle is identified. The first sensor data is defined in both a global coordinate system and a local coordinate system for the first vehicle. A second vehicle is identified based on a second location of the second vehicle within a threshold distance of the first vehicle within the first timeframe. The second vehicle is associated with second sensor data that is further associated with a label identifying a location of an object, and the location of the object is defined in a local coordinate system of the second vehicle. A conversion from the local coordinate system of the second vehicle to the local coordinate system of the first vehicle may be determined and used to transfer the label from the second sensor data to the first sensor data.

Abstract: Example embodiments relate to self-reflection filtering techniques within radar data. A computing device may use radar data to determine a first radar representation that conveys information about surfaces in a vehicle's environment. The computing device may use a predefined model to generate a second radar representation that assigns predicted self-reflection values to respective locations of the environment based on the information about the surfaces conveyed by the first radar representation. The predefined model can enable a predefined self-reflection value to be assigned to a first location based on information about a surface positioned at a second location and a relationship between the first location and the second location. The computing device may then modify the first radar representation based on the predicted self-reflection values in the second radar representation and provide instructions to a control system of the vehicle based on modifying the first radar representation.

Abstract: Aspects and implementations of the present disclosure address shortcomings of the existing technology by enabling lidar-assisted segmentation and identification of particulate matter in autonomous vehicle (AV) applications, by: obtaining, by a sensing system of the AV, a plurality of return points, each return point having one or more velocity values and one or more coordinates of a reflecting region that reflects a signal emitted by the sensing system, identifying, in view of the one or more velocity values of each of a first set of the return points of the plurality of return points, that the first set of the return points is associated with a particulate matter in an environment of the AV, and causing a driving path of the AV to be determined in view of the particulate matter.

Abstract: The technology relates to developing a highly accurate understanding of a vehicle's sensor fields of view in relation to the vehicle itself. A training phase is employed to gather sensor data in various situations and scenarios, and a modeling phase takes such information and identifies self-returns and other signals that should either be excluded from analysis during real-time driving or accounted for to avoid false positives. The result is a sensor field of view model for a particular vehicle, which can be extended to other similar makes and models of that vehicle. This approach enables a vehicle to determine when sensor data is of the vehicle or something else. As a result, the detailed modeling allowing the on-board computing system to make driving decisions and take other actions based on accurate sensor information.

Abstract: One example method involves rotating a housing of a light detection and ranging (LIDAR) device about a first axis. The housing includes a first optical window and a second optical window. The method also involves transmitting a first plurality of light pulses through the first optical window to obtain a first scan of a field-of-view (FOV) of the LIDAR device. The method also involves transmitting a second plurality of light pulses through the second optical window to obtain a second scan of the FOV. The method also involves identifying, based on the first scan and the second scan, a portion of the FOV that is at least partially occluded by an occlusion.

Abstract: Example implementations may relate to determining a strategy for a drop process associated with a light detection and ranging (LIDAR) device. In particular, the LIDAR device could emit light pulses and detect return light pulses, and could generate a set of data points representative of the detected return light pulses. The drop process could involve a computing system discarding data point(s) of the set and/or preventing emission of light pulse(s) by the LIDAR device. Accordingly, the computing system could detect a trigger to engage in the drop process, and may responsively (i) use information associated with the environment around the vehicle, operation of the vehicle, and/or operation of the LIDAR device as a basis to determine the strategy for the drop process, and (ii) engage in the drop process in accordance with the determined strategy.

Abstract: Example embodiments relate to radar image video compression techniques using per-pixel Doppler measurements, which can involve initially receiving radar data from a radar unit to generate a radar representation that represents surfaces in the environment. Based on Doppler scores in the radar representation, a range rate can be determined for each pixel that indicates a radial direction motion for a surface represented by the pixel. The range rates and backscatter values can then be used to estimate a radar representation prediction for subsequent radar data received from the radar unit, which enables a generation of a compressed radar data file that represents the difference between the radar representation prediction and the actual representation determined for the subsequent radar data. The compressed radar data file can be stored in memory, transmitted to other devices, and decompressed and used to train models via machine learning.

Abstract: The present disclosure relates to systems and methods for occlusion detection. One example method involves a light detection and ranging (LIDAR) device scanning at least a portion of an external structure within a field-of-view (FOV) of the LIDAR device. The LIDAR device is physically coupled to the external structure. The scanning comprises transmitting light pulses toward the external structure through an optical window, and receiving reflected light pulses through the optical window. The reflected light pulses comprise reflections of the transmitted light pulses returning back to the LIDAR device from the external structure. The method also involves detecting presence of an occlusion that at least partially occludes the LIDAR device from scanning the FOV based on at least the scan of the at least portion of the external structure.

Abstract: Aspects and implementations of the present disclosure address shortcomings of the existing technology by enabling efficient object identification and tracking in autonomous vehicle (AV) applications by using velocity data-assisted mapping of first set of points obtained for a first sensing data frame by a sensing system of the AV to a second set of points obtained for a second sensing data frame by the sensing system of the AV, the first set of points and the second set of points corresponding to an object in an environment of the AV, and causing a driving path of the AV to be determined in view of the performed mapping.

Abstract: Example implementations may relate to determining a strategy for a drop process associated with a light detection and ranging (LIDAR) device. In particular, the LIDAR device could emit light pulses and detect return light pulses, and could generate a set of data points representative of the detected return light pulses. The drop process could involve a computing system discarding data point(s) of the set and/or preventing emission of light pulse(s) by the LIDAR device. Accordingly, the computing system could detect a trigger to engage in the drop process, and may responsively (i) use information associated with the environment around the vehicle, operation of the vehicle, and/or operation of the LIDAR device as a basis to determine the strategy for the drop process, and (ii) engage in the drop process in accordance with the determined strategy.

Abstract: Example embodiments relate to self-reflection filtering techniques within radar data. A computing device may use radar data to determine a first radar representation that conveys information about surfaces in a vehicle's environment. The computing device may use a predefined model to generate a second radar representation that assigns predicted self-reflection values to respective locations of the environment based on the information about the surfaces conveyed by the first radar representation. The predefined model can enable a predefined self-reflection value to be assigned to a first location based on information about a surface positioned at a second location and a relationship between the first location and the second location. The computing device may then modify the first radar representation based on the predicted self-reflection values in the second radar representation and provide instructions to a control system of the vehicle based on modifying the first radar representation.

Abstract: The technology relates to determining a source of a horn honk or other noise in an environment around one or more self-driving vehicles. Aspects of the technology leverage real-time information from a group of self-driving vehicles regarding received acoustical information. The location and pose of each self-driving vehicle in the group, along with the precise arrangement of acoustical sensors on each vehicle, can be used to triangulate or otherwise identify the actual location in the environment for the origin of the horn honk or other sound. Other sensor information, map data, and additional data can be used narrow down or refine the location of a likely noise source. Once the location and source of the noise is known, each self-driving vehicle can use that information to modify current driving operations and/or use it as part of a reinforcement learning approach for future driving situations.

Abstract: The subject matter of this specification can be implemented in, among other things, a system that includes a light source to produce a first beam and one or more first optical elements to impart an orbital angular momentum (OAM) to at least some of a plurality of output beams obtained from the first beam. The system further includes an output optical device to direct the output beams towards a target object and a plurality of photodetectors to generate signals representative of a difference between an input phase information, carried by one or more beams reflected from the target object, and an output phase information, carried by one of local copies of the output beams. The system further includes a processing device to determine, using the difference between the input phase information and the output phase information, one or more orthogonal components of a velocity of the target object.

Abstract: Aspects and implementations of the present disclosure address shortcomings of the existing technology by enabling Doppler-assisted segmentation of points in a point cloud for efficient object identification and tracking in autonomous vehicle (AV) applications, by: obtaining, by a sensing system of the AV, a plurality of return points comprising one or more velocity values and one or more coordinates of a reflecting region that reflects a signal emitted by the sensing system, the one or more velocity values and the one or more coordinates obtained for the same instance of time, identifying that the set of the return points is associated with an object in an environment, and causing a driving path of the AV to be determined in view of the object.