Abstract: Techniques for performing prediction based on relative lateral and/or longitudinal position data are described herein. A vehicle may detect an object in an environment. The vehicle may generate an embedding associated with the object and input the embedding into a machine-learned model. The machine-learned model may generate a rotation matrix based on the lateral or longitudinal of object(s), and apply the rotation matrix to the embedding. Based on modifying the embedding (or generating a modified embedding), the attention layer of the machine-learned model may perform attention on the modified embedding and respond by outputting output data. The vehicle may rotate the output data (or generate rotated output data) which may be used by one or more machine-learned models to predict object behavior, generate vehicle actions, etc.

Abstract: A reflector unit may behave as a retroreflector to reflect light of a selected color. The reflector unit may comprise a plurality of reflector elements having a first side configured to reflect light of a first color and a second side configured to reflect light of a second color. In response to receiving a signal (e.g., indicating a bi-directional vehicle has changed a direction of travel), the reflector unit may cause the reflector elements to rotate and expose a different side of the reflector elements depending on the direction of travel.

Abstract: Techniques are provided comprising receiving map data and determining a first polygon representing a first traversable surface and a second polygon representing a second traversable surface. The first polygon may include a first portion that overlaps with a second portion of the second polygon. The techniques may determine, based at least in part on elevation information associated with the first traversable surface and the second traversable surface, respective elevation layers associated with the first polygon and the second polygon. It may be determined whether to merge the first polygon and the second polygon based at least in part on whether the first polygon is associated with a same elevation layer as the second polygon. Based at least in part on determining whether to merge the first and second polygons, an updated map may be determined.

Abstract: Techniques are discussed herein for training and executing machine learning (ML) prediction models used to control autonomous vehicles in driving environments. In various examples, ML prediction models configured to output joint trajectory predictions for multiple objects in an environment may be trained by evaluating the interactions between the objects represented by the predicted trajectories. A training component may train an ML prediction model using a standard loss function based on the accuracy of the predicted trajectories relative to the ground truth trajectories, and based on an auxiliary loss determined by the agent-to-agent interactions represented by the predicted trajectories.

Abstract: The disclosure describes systems, as well as methods and computer-readable media for deploying a ramp element of a vehicle. A first actuator assembly is provided to releasably couple the ramp element to a first portion of the vehicle and a second actuator assembly is provided to rotate the ramp element around a first axis, arranged at a proximal portion of the ramp element, to move the ramp element from a stowed position to a deployed position.

Abstract: Techniques are discussed for determining prediction probabilities of an object based on a top-down representation of an environment. Data representing objects in an environment can be captured. Aspects of the environment can be represented as map data. A multi-channel image representing a top-down view of object(s) in the environment can be generated based on the data representing the objects and map data. The multi-channel image can be used to train a machine learned model by minimizing an error between predictions from the machine learned model and a captured trajectory associated with the object. Once trained, the machine learned model can be used to generate prediction probabilities of objects in an environment, and the vehicle can be controlled based on such prediction probabilities.

Abstract: Techniques for determining detection boxes representing objects in an environment using improved corner detections are disclosed herein. Autonomous vehicle sensors can capture data in an environment that may include separate objects, such as large and/or articulated vehicles. In an example, the data can include a plurality of points that may be processed to identify object corners. Locations of points in clusters of points associated with particular object corners may be aggregated to determine a predicted corner point that may be used to adjust the detection box for that object, thereby generating a more accurate detection box for the object. A vehicle computing system can control the vehicle using the adjusted detection box.

Abstract: Sensors may be used to detect objects in an environment. In an example, a vehicle may include a radar sensor and one or more additional sensors. The additional sensors generate sensor data used to detect objects in the environment and determine attributes associated with the detected objects. Radar data from the radar sensor is compared to the attributes of the detected object to identify radar returns that are associated with the object. Object data is generated that associates velocity information for the identified radar returns with the object.

Abstract: Techniques for determining an object velocity associated with a radar device are discussed herein. A computing device can implement a model that receives radar data from the radar device as input and identifies a Doppler interval for determining candidate correction values. The model can select one of the candidate correction values for processing when determining the object velocity. For example, the model can determine a Doppler velocity for output by a radar device based on the selected correction value. The object velocity can be used by a vehicle computing device for predicting vehicle actions to control a vehicle.

Abstract: Techniques include generating a driving scenario representation based on another driving scenario representation and a modification representation. In examples, a method includes receiving a first driving scenario; receiving a second driving scenario; providing the first driving scenario and the second driving scenario to a first model; receiving, from the first model, a first representation associated with the first driving scenario and a second representation associated with the second driving scenario; determining a modification representation representing a difference measure associated with the first representation and the second representation; training the first model based at least in part on the modification representation; and providing the first model to a second system. The second system may be configured to modify a third driving scenario based at least in part on the first model to determine a fourth driving scenario, and control a vehicle based at least in part the fourth driving scenario.

Abstract: A method comprises (i) obtaining, by a first sensor of a plurality of sensors, first sensor data associated with one or objects in an environment in which one or more vehicles are located, (ii) obtaining, by a second sensor, second sensor data associated with one or more objects, (iii) determining contact metrics associated with the one or more objects associated with the first sensor data and the one or more objects associated with the second sensor data, (iv) determining one or more properties of one or more network connections to a remote system, (v) determining, based at least in part on the contact metrics and the one or more properties, a subset of sensor data to send to the system comprising at least some of the first sensor data and/or at least some of the second sensor data, and (vi) sending the subset of the sensor data to the system.

Abstract: Techniques are discussed for generating, maintaining, and accessing maps to be used for vehicle localization and navigation. A map of a region may include map data representing discrete tiles of the region, and be associated with a map manifest listing pointers to content of the map. In some cases, the map manifest may have a recursive structure which references other map manifests. The map files may be stored on a storage system that avoids duplication and allows efficient determination of changes to content. For example, the stored map files may each be associated with a content-based hash value, which may be incorporated into a filename or URI of the map file. Computing systems on a vehicle may use comparison of the content-based hash values to determine files that have changed and need to be downloaded. The computing systems may also use hash values to determine accuracy of downloaded files.

Abstract: A system and method for selecting a trajectory to traverse comprises receiving a primary trajectory, determining, based at least in part on the primary trajectory, a primary area traversed by the sensor platform in accordance with the primary trajectory and a secondary area traversed by the sensor platform in accordance with a contingent trajectory, and determining, based at least in part on a sensor associated with the sensor platform, that an object is located at least partly within a combined area comprising the primary and secondary area. Upon determining that an object is in the combined area, determining, based at least in part on the object being located at least partly within the combined area, a probability of a collision with the object; and controlling, based at least in part on the determined probability of collision, the sensor platform.

Abstract: Techniques for determining objective map quality and evaluating updated maps against existing maps to determine an operational map for use in controlling a vehicle in an environment are disclosed. A map and map data for an environment may be determined based on collected sensor data. A variety of map quality indicators may be determined based on such data and used to determine an overall quality of the map. The overall quantity for the map may be compared to a current operational map for the environment. The operational map may be replaced with the newer map if the newer map has a higher map quality. Otherwise, the current operational map may be retained. The map quality measurements described herein may also be used to determine whether changes in map generation processes and operations result in improved quality maps.

Abstract: Techniques for enabling a vehicle to navigate around a blocking object, such as double-parked vehicle (“DPV”), are discussed herein. In some examples, a vehicle may receive sensor data representative of an environment. The vehicle may analyze such sensor data to determine that a DPV is located proximate the vehicle. When determining whether to follow the trajectory, the vehicle may determine one or more cost values corresponding to the trajectory. Further, the vehicle may receive weighted heatmap(s) that cover the region proximate the DPV. The vehicle may modify some or all cost values of the trajectory based on the weighted values of the heatmap(s). In some examples, the vehicle may follow the trajectory based at least in part on the cost values.

Abstract: Techniques are discussed herein for training and executing movement and navigation models for controlling autonomous vehicles in simulations and/or in real-world physical environments. In some examples, a model training component may receive object movement data, such as velocities and/or steering angles, from vehicle logs or other ground truth data sources. The movement data may be discretized into labeled bins, and a smoothing function (e.g., a Gaussian quadrature or other weight sum function) centered on the ground truth data may be applied to the labeled bins. During model training, the model outputs can be compared to the smoothed object movement data using a loss function (e.g., a cross-entropy loss function), to reduce the magnitude of the loss penalties for incorrect predictions that are closer to the ground truth data. As a result, models trained using the techniques described herein may provide improved object movement predictions and/or navigation performance for controlling autonomous vehicles.

Abstract: Techniques for enabling operation of a vehicle in one or more different operational states after detecting a fault associated with a component of the vehicle are described herein. A vehicle computing system can monitor components of the vehicle and identify a fault associated with a component. Based on the fault, the vehicle computing system can identify a degraded operational state for the vehicle including one or more limitations or constraints on operation of the vehicle, such as limited speed, acceleration, direction of travel, and/or the like. The vehicle computing system can additionally determine a threshold time period, distance, or range for operating the vehicle in the degraded operational state. The vehicle computing system can cause the vehicle to be controlled according to the limitation(s) or constraint(s) for a time period, distance, or range that is equal to or less than the associated threshold.

Abstract: A method is provided comprising receiving log data associated with a vehicle traversing a real-world environment, and inputting the log data into a search algorithm. A first modified value of a first parameter associated with the log data is received from the search algorithm. The search algorithm determines the first modified value based on a first simulation score associated with occurrence of a first event in a first run of a driving simulation. The first run of the driving simulation includes an autonomous vehicle and is based at least in part on the first modified value of the first parameter. The first modified value of the first parameter is output, and may be used for validating or testing a machine learning model such as a machine learning model for controlling an autonomous vehicle.

Abstract: Techniques for improving scenario determinations for generate testing scenarios for vehicle control system testing are disclosed. A model generates scenario embeddings based on a natural language scenario description. Similar scenarios are identified and/or generated based on similarities between the embeddings for such scenarios and the embeddings determined for the scenario description. The similar scenarios are evaluated for similarity quality and the results of this evaluation are used to adjust the embeddings, embedding determination process, and/or the similarity determination process. This process may be iteratively performed until the system consistently generates and/or identifies sufficiently similar scenarios.