Abstract: Techniques for reducing stray light or glare in sensor data are discussed herein. The sensor data can be captured by a sensor, and the sensor data can be based on laser pulses emitted from emitters associated with a first polarization and received by the sensor associated with a second polarization. A difference between the first polarization and the second polarization can be used to filter out or reduce stray light from the sensor and sensor data. The sensor data can be utilized to determine distance data associated with an object. The sensor data and the first polarization and second polarization can also be utilized to determine characteristics of the object.

Abstract: Techniques are described herein for determining whether multipath reflections are present in a sensed signal associated with a time-of-flight sensor. In various examples, multiple signals may be generated associated with differing wavelengths and exposure times. Such signals may be used to reference a map of values generated from, in some examples, simulations of environments in which multipath reflections are likely. The map, in turn, may provide one or more values associated with one or more distances associated with an object and a multipath reflection. Additionally, an emitted waveform may form the basis of a reference signal for demodulation of the plurality of signals to account for manufacturing, thermal, and other defects. The resultant distances may be used to filter data and remove “bad” pixels to form a resultant depth-image which may be used, for example, to control a vehicle.

Abstract: Techniques for generating (e.g., creating and/or updating) map data associated with an environment based on sensor data. An example system may determine (e.g., based on the sensor data) and/or receive an indication of a lane boundary associated with a path. The system may determine, based on a set of drivable areas associated with the path, a polygon. The system may determine a representative line associated with the polygon and determine a set of points along the representative line. The system may determine a set of intersections of a set of lines that originate at the set of points and extend laterally from the representative line. The system may determine, based on the set of intersections, a representation of the lane boundary and associate the representation with the map data.

Abstract: In order to operate reliably and effectively, an autonomous vehicle may at least partially rely on accurately determining an initial position and orientation of the vehicle within the environment. Without an accurate localization of the vehicle's location, position, and orientation in an environment, the vehicle's perception and/or planning components may be hindered, thereby reducing the effectiveness and/or safety of the vehicle. Techniques of the disclosure herein are related to initializing a localization component of a vehicle.

Abstract: Techniques are disclosed in which sensor data may be received from a sensor associated with a vehicle in an environment. Based at least in part on the sensor data, a set of reachable positions of a first object in the environment in a future time window may be determined. Based at least in part on the set of reachable positions, a first metric may be determined. The first metric may be associated with likelihood of the first object interacting with the vehicle in the future time window. It may be determined, based at least in part on the first metric, to predict a future position of the first object based on a first or a second prediction method. Techniques may determine a trajectory for controlling the vehicle based at least in part on the future position of the first object.

Abstract: Lidar sensor alignment validation may ensure that a relative position and/or orientation of two or more lidar sensors is accurate by determining a sensor alignment error associated with lidar data received from two or more lidar sensors. In some examples, the sensor alignment error may be determined based at least in part on a median angular error and/or median angle determined based at least in part on the lidar data. The median angular error and/or median angle may be determined based at least in part on a first distance from a first lidar point of a first lidar device to a first geometric approximation of a second lidar device's lidar data within a voxel and/or a second distance from a second lidar point of the second lidar device to a second geometric approximation of the first lidar device's lidar data within the voxel.

Abstract: Techniques for determining a match prediction associated with image data and map data associated with a vehicle environment are described herein. In some cases, the techniques described herein enable determining the match prediction based on: (i) an image segment that represents at least a portion of a lane marking as detected based on the image data, and (ii) a map segment that represents at least a portion of a lane marking as determined based on the map data. In some cases, to determine the match prediction, an example system determines: (i) a set of first errors between map points associated with the map segment and the image segment, and (ii) a set of second errors between image points associated with the image segment and the map segment.

Abstract: Techniques for engaging and disengaging a vehicle from an autonomous operation mode are described. The techniques include receiving a request to engage or disengage the autonomous mode at a teleoperations system. The request is validated at the teleoperations system using a first set of conditions describing trust in the request and validity of the request. The request is then sent, if valid, to the vehicle where a second validation is performed using a second set of conditions describing the conditions or states of the vehicle. If the second conditions are valid then the request is performed and the vehicle engages or disengages from autonomous mode.

Abstract: Techniques for requesting remote assistance from a vehicle are discussed. The techniques include receiving, from a remote computing device, coordinates of a footprint in which the vehicle is capable of stopping. The techniques further include receiving, from the remote computing device, a target orientation associated with the footprint. The techniques may determine a path to the footprint based to achieve target orientation associated with the footprint.

Abstract: A seat of a vehicle is designed for safety to reduce risk of injury to an occupant of the seat in the event of an abrupt movement or a collision involving the vehicle from the rear. The seat may have seatback geometry designed to promote coupling an occupant to the seat, thereby minimizing a gap between an occupant and the seatback and, therefore, the impact forces to the occupant in a rear-facing collision by reducing a relative velocity between the occupant and the seat, as well as absorbing energy of the occupant over a longer distance. The seat may include one or more load plates and materials that plastically deform under a force imparted on the seat by the occupant accelerating relative to and toward the seat during a vehicle collision. The load plates distribute the force across the energy absorbing material to reduce the maximum force experienced by the occupant.

Abstract: Techniques are described herein for converting driving logs to driving simulations and for analysis and improvement of said conversion. A driving log relating to a driving scenario may be received and a feature may be determined. A driving simulation of the driving simulation may be generated based on the driving log, and a value may be determined that indicates an ability of a system generating the simulation to represent the feature. The value may be stored. The driving simulation may also be stored.

Abstract: Data-driven techniques are described herein for determining collision probabilities and/or predicted collision rates associated with vehicle control components used in autonomous vehicles. A performance modeling system may determine a first vehicle-distance-to-object distribution, based on log data captured in real-world driving environments. A second tracking error distribution may be received and/or determined associated with one or more vehicle control components, and the performance modeling system may determine collision probabilities and/or rates associated with the vehicle control components based on the distributions. In various examples, these techniques can be used to determine target error distributions associated with particular vehicle control components and/or particular driving maneuvers, based on overall safety and performance target metrics for an autonomous vehicle.

Abstract: Systems and methods for disengaging or engaging autonomy remotely are provided. The autonomous vehicle may receive a request from a remote computing system associated with a remote operator. The request may require the autonomous vehicle to disengage, engage, or re-engage autonomy. The autonomous vehicle may determine whether the request is valid, and whether certain safety criteria are met. Upon determining that the request is valid and safety criteria are met, the autonomous vehicle may process the request and cause the autonomous vehicle to enter the desired state. The autonomous vehicle may send a message to the remote computing system, indicating whether or not the autonomous vehicle has entered the desired state.

Abstract: Systems and techniques for determining deceleration controls to use in a trajectory for use in stopping a vehicle are described. A vehicle control system may determine if a stopping point is represented in a spatially-generated trajectory (e.g., controls to be commanded at particular locations) and, if so, whether the points in the trajectory approaching the stopping point meet near-stop conditions. If so, the system operated the vehicle in stop vehicle control mode, generating temporally-based controls to slow the vehicle more comfortably. Otherwise, the system operates the vehicle as normal, using spatially-based controls for controlling the vehicle.

Abstract: Systems and methods may determine whether contagion(s) are being spread within a vehicle using sound and/or other sensor data. The vehicle may include sensors, such as microphones, cameras, and the like, that continuously or periodically monitor sound and other conditions in an interior of the vehicle and generate sensor data for use in determining whether contagion(s) have been introduced into the vehicle. In the case of sound, audio data generated by the microphones, for example, may represent acoustic characteristic(s) associated with the sound(s) produced by the occupant(s). The introduction of contagion(s) may be used when determining whether to perform action(s) associated with sanitizing the vehicle and/or otherwise reducing a spread of viruses, bacteria, and/or germs among the occupant(s) and/or within the vehicle.

Abstract: Generating a semantic segmentation of point cloud data may include detecting an object based at least in part one or more images and using the image-based object detection in combination with the point cloud data to determine the semantic segmentation of the point cloud data. The semantic segmentation of point cloud data may be conducted in parallel with executing an object tracking component. Parallelizing these functions reduces the amount of time it takes to use sensors to detect and track objects and the point cloud semantic segmentation may provide further granularity for planning and ML model training.

Abstract: Mitigating the impact of cloud service downtime, migration to a new cloud instance/container, and/or software update on a vehicle or fleet of vehicles may include receiving an indication that a vehicle is experiencing or is scheduled to experience downtime; determining, based at least in part on the indication, a criterion indicating a minimum vehicle state for the vehicle to be prepared for the downtime; determining, based at least in part on sensor data received by the vehicle, a vehicle state associated with the vehicle; and causing control of the vehicle based at least in part on the criterion and the vehicle state data.

Abstract: Systems and techniques for determining a buffer region for use in controlling a vehicle and avoiding collisions are disclosed herein. A predicted region of travel of a vehicle front bumper may be determined. The position of the front bumper may be determined at points along a center curve of the predicted region of travel and polygons may be determined for the positions. The polygons may be joined and modified using a convex shape-based algorithm to determine a convex polygonal buffer region that is used in collision detection.

Abstract: Techniques for generating image data of a modality different from a visible light modality are discussed herein. For example, visible light images may be readily available for training and testing vehicle systems, but corresponding image data of a thermal modality may not be available. In examples, one or more trained machine learning (ML) models may be used to generate synthetic thermal images based on an input visible light image. Such ML models may be trained, based on a training dataset including metadata and semantic segmentation associated with visible light images, to determine an appearance of the output synthetic thermal images. In some examples, an emittance level of regions of the synthetic thermal image may be further adjusted based on information associated with the respective region. The synthetic thermal images may be used for simulation and training of vehicle systems to interpret thermal imagery.