Abstract: Sensor data obtained from vehicles driving through a particular environment (e.g., a particular city or region being mapped) is used to identify and label traffic control features. Location, speed, and/or acceleration of mapping vehicles can be used to identify intersections that may have traffic control features. Environmental data, such as image data captured by one or more cameras, and point cloud data collected by lidar and/or radar sensors, is used to automatically detect traffic control features at the intersections.

Abstract: Planning, or path finding is one of many tasks performed by an autonomous vehicle (AV). Planning involves searching for and updating an optimal plan from a current pose to an end pose while avoiding obstacles. Planning can be particularly challenging in driving scenarios involving complex driving maneuvers, such as parking, making a U-turn, making a K-turn, etc. Searching for an optimal plan may take more time, e.g., a few seconds, in certain driving scenarios. Consumers of the plan may have a low tolerance for delay in receiving the optimal plan. A relay can be implemented to publish the latest plan to the consumers at a frequency. The relay can check whether the delay violates a dynamic threshold and raise an error accordingly. The searching process can be improved to reduce the delay in finding an optimal plan.

Abstract: The subject technology is related to autonomous vehicles (AV) and, in particular, to calibration of an sensors of an AV on open roads after calibrating the sensors in an undefined training area. An example method includes calibrating a plurality of sensors of the AV based on performing a plurality of training steps of increasing complexity within the undefined environment, wherein the plurality of sensors of the AV are uncalibrated with respect to each other, determining that the AV satisfies an open road requirement to navigate an open road based on completing the plurality of training steps in the undefined environment, and operating the AV on open roads subject to at least one constraint. The method of claim 1, wherein the at least one constraint comprises a list of prohibited maneuvers that limit operation of the AV on open roads. The constraints are configured to dynamically change.

Abstract: Systems and methods for a vehicle to use a fallback control system orthogonal to a primary control system are provided. A method includes collecting, by a fallback control system of a vehicle, sensor data generated by one or more sensors of the vehicle; determining, by the fallback control system using a first process different from a second process of a primary control system of the vehicle, an alternative action for the vehicle based on the sensor data; and sending, by the fallback control system based on the alternative action, an instruction for modifying an action associated with the primary control system upon a faulty condition associated with the primary control system.

Abstract: Systems and methods for collision imminent detection are provided. A method includes collecting sensor data generated by one or more sensors of a vehicle; detecting, by a fallback control system of the vehicle based on the sensor data, an imminent collision between the vehicle and an object in an environment of the vehicle using a first process different from a second process of a primary control system of the vehicle, wherein the first process and the second process are associated with at least one of a perception or a prediction; and transmitting, by the fallback control system to the primary control system, an indication of the detected imminent collision.

Abstract: System, methods, and computer-readable media for a calibration check process of sensors on an autonomous vehicle (AV) via a drive-through environment mapped with a series of targets. The targets are used to check the calibration of the sensors on the AV as a quality check before the AV is determined fit to drive autonomously. Supervising a calibration process that collects data from the sensors on the AV based on exposure to a series of targets or augmented camera targets. The collected data may be used for extrinsic calibration aimed to determine that extrinsic parameters that define the rigid relationship between sensors are in set checker bounds.

Abstract: It may be desirable for a remote advisor to request an autonomous vehicle to relocate to a target relocation pose using a freespace planner that has fewer constraints than the structured planner that is used for normal driving. The autonomous vehicle planning stack can include a remote vehicle interface, which can serve as the application programming interface for remote requestors to request relocation of the autonomous vehicle. The remote vehicle interface may implement a relocation state machine to carry out different parts of a relocation session. A scenario manager may subscribe to the state of the relocation state machine. Based on the state, the scenario manager can request the freespace planner to generate an output plan for the target relocation pose and can orchestrate handoff between different motion planners to relocate the autonomous vehicle.

Abstract: It may be desirable for a remote advisor to request an AV to relocate to a target relocation pose. In some cases, reaching a target relocation pose may involve a multi-point maneuver. Ability to perform relocation that involves a multi-point maneuver offers flexibility for the AV to maneuver around double-parked vehicles, to avoid obstacles, to aid a reroute when a route becomes unnavigable, etc. To support relocation that involves performing multi-point maneuvers, the planning stack may be modified to include a motion planner that can generate output plans that involve the AV going forward and in reverse. Moreover, the vehicle control stack may publish vehicle controls information as feedback information to the motion planner and the remote requestor, so that the motion planner and the remote requestor can be informed of the state of the AV.

Abstract: The disclosed technology provides solutions for updating high definition maps based on low resolution map assets. In some aspects, a process of receiving a change detection relating to a change in the real world is provided. The process can include steps for receiving autonomous vehicle drive data based on the change in the real world, generating low resolution tile data based on the autonomous vehicle drive data based on the change in the real world, generating updated semantic data based on the low resolution tile data generated, and providing the updated semantic data to an autonomous vehicle to update a proximate area of the change in the real world of a base map of the autonomous vehicle. Systems and machine-readable media are also provided.

Abstract: A method is described and includes receiving an indication that a collision between a vehicle and an object is predicted; processing first sensor data obtained from a first onboard sensor of the vehicle to determine whether an actual occurrence of the predicted collision can be confirmed from the first sensor data and if not, processing second sensor data obtained from a second onboard sensor of the vehicle to determine whether the actual occurrence of the predicted collision can be confirmed from the second sensor data. If the actual occurrence of the predicted collision is confirmed from one of the first sensor data and the second sensor data, determining a severity of the collision based on the indication and the one of the first sensor data and the second sensor data; and prompting an action to be based on the determined severity of the collision.

Abstract: Systems and techniques are provided for initializing the localization of an autonomous vehicle (AV) based on sensor data. An example method can include determining a first position of an AV prior to a power cycle, wherein the power cycle begins when the AV is powered off and ends when the AV is powered on; receiving, after the power cycle, sensor data from one or more sensors of the AV, the sensor data comprising image data; and in response to a determination that the AV has completed the power cycle, determining, based on image data from one or more sensors of the AV, whether a second position of the AV after the power cycle matches the first position of the AV prior to the power cycle.

Abstract: The present technology provides systems, methods, and devices that can dynamically augment aspects of a map as an autonomous vehicle navigates a route, and therefore avoids the need for dispatching a special purpose mapping vehicle to keep navigating a route. As the autonomous vehicle navigates a route, the autonomous vehicle can determine that current data captured by at least one sensor of the autonomous vehicle describing a location is inconsistent with the primary map of the location. The autonomous vehicle can determine that a portion of the current data describes a second feature that is distinct from a first feature described by a primary map. The second feature can be added to an augmented map that is based on the primary map, and the position of the autonomous vehicle can be located with respect to the first feature rather than the second feature on the augmented map.

Abstract: A method is described and includes receiving a frame of point cloud data from at least one onboard light detection and ranging (LIDAR) sensor of a vehicle; discarding points of the received frame of point cloud data that are outside a defined geographic area around the vehicle; and, subsequent to the discarding, performing ray tracing in connection with the remaining points of the received frame of point cloud data. The method may further include characterizing an occupancy condition of each of a plurality of cells of a three-dimensional (3D) grid corresponding to the defined geographic area based on the ray tracing of the received frame of point cloud data, wherein the 3D grid corresponds to the received frame of point cloud data.

Abstract: Systems and methods are disclosed for detecting crosswalk locations. Crosswalk locations can be detected by determining at least two painted lines on a road surface, and then combining the at least two painted lines on the road surface into a grouping of related elements. The grouping of related elements can be classified into a crosswalk (or in some embodiments a type of crosswalk) based on attributes of the grouping of related elements, where the attributes of the grouping of the elements includes a distance between the at least two painted lines, and an orientation of the grouping of related elements on the road surface.

Abstract: Systems, methods, and computer-readable media are provided for detecting a pavement marking around an autonomous vehicle, comparing the detected pavement marking with a pavement marking present in a semantic data map, determining whether a change has occurred between the detected pavement marking and the pavement marking present in the semantic data map, and updating the semantic data map based on the determining of whether the change has occurred between the detected pavement marking and the pavement marking present in the semantic data map.

Abstract: System, methods, and computer-readable media for a calibration check process of sensors on an autonomous vehicle (AV) via a drive-through environment mapped with a series of targets. The targets are used to check the calibration of the sensors on the AV as a quality check before the AV is determined fit to drive autonomously. Supervising a calibration process that collects data from the sensors on the AV based on exposure to a series of targets or augmented camera targets. The collected data may be used for extrinsic calibration aimed to determine that extrinsic parameters that define the rigid relationship between sensors are in set checker bounds.

Abstract: The subject technology is related to autonomous vehicles (AV) and, in particular, to calibration of an sensors of an AV on open roads after calibrating the sensors in an undefined training area. An example method includes calibrating a plurality of sensors of the AV based on performing a plurality of training steps of increasing complexity within the undefined environment, wherein the plurality of sensors of the AV are uncalibrated with respect to each other, determining that the AV satisfies an open road requirement to navigate an open road based on completing the plurality of training steps in the undefined environment, and operating the AV on open roads subject to at least one constraint. The method of claim 1, wherein the at least one constraint comprises a list of prohibited maneuvers that limit operation of the AV on open roads. The constraints are configured to dynamically change.

Abstract: The subject technology is related to autonomous vehicles (AV) and, in particular, to calibrating multiple sensors of an AV in an undefined training area. An example method includes instructing the AV to pilot itself in an undefined training area subject to at least one constraint, wherein the AV is instructed to pilot itself along a path until the AV has overlapped at least a portion of the path, and at a location at which the AV has overlapped at least the portion of the path, determining that first returns from a previous LIDAR scan overlaps with second returns from a subsequent LIDAR scan taken when the AV has overlapped at least the portion of the path. Initially, the AV does not include a location reference to identify locations of objects in the undefined training area and a plurality of sensors of the AV are uncalibrated with respect to each other.

Abstract: A method is described and includes receiving an indication that a collision between a vehicle and an object is predicted; processing first sensor data obtained from a first onboard sensor of the vehicle to determine whether an actual occurrence of the predicted collision can be confirmed from the first sensor data and if not, processing second sensor data obtained from a second onboard sensor of the vehicle to determine whether the actual occurrence of the predicted collision can be confirmed from the second sensor data. If the actual occurrence of the predicted collision is confirmed from one of the first sensor data and the second sensor data, determining a severity of the collision based on the indication and the one of the first sensor data and the second sensor data; and prompting an action to be based on the determined severity of the collision.

Abstract: A method is described and includes receiving a frame of point cloud data from at least one onboard light detection and ranging (LIDAR) sensor of a vehicle; discarding points of the received frame of point cloud data that are outside a defined geographic area around the vehicle; and, subsequent to the discarding, performing ray tracing in connection with the remaining points of the received frame of point cloud data. The method may further include characterizing an occupancy condition of each of a plurality of cells of a three-dimensional (3D) grid corresponding to the defined geographic area based on the ray tracing of the received frame of point cloud data, wherein the 3D grid corresponds to the received frame of point cloud data.