Abstract: Techniques for determining a response of a simulated vehicle to a simulated object in a simulation are discussed herein. Log data captured by a physical vehicle in an environment can be received. Object data representing an object in the log data can be used to instantiate a simulated object in a simulation to determine a response of a simulated vehicle to the simulated object. Additionally, one or more trajectory segments in a trajectory library representing the log data can be determined and instantiated as a trajectory of the simulated object in order to increase the accuracy and realism of the simulation.

Abstract: Techniques for performing computer vision operations for a vehicle using image data of an environment of the vehicle are described herein. In some cases, a vehicle (e.g., an autonomous vehicle) can determine a predicted position (including a predicted depth) and/or a predicted trajectory for an object in the vehicle environment based on data generated by projecting a map object described by the map data for a vehicle environment to image data of the vehicle environment.

Abstract: Techniques for determining a parking trajectory for a vehicle are discussed herein. A parking management component may determine or receive a three-dimensional grid (“grid”), discretized based at least in part on a heading offset, a lateral offset, and/or a longitudinal offset between a first and second pose. A cell of the grid may include a cost associated indicating a minimum difference to the second pose when moving from the first as may be limited based on kinematic and/or dynamic constraints. When driving, a vehicle may determine a relative state of the vehicle to a desired location, use the relative state to access a cost from the grid, and determine whether to follow a trajectory based on the cost.

Abstract: Techniques for representing a scene or map based on statistical data of captured environmental data are discussed herein. In some cases, the data (such as covariance data, mean data, or the like) may be stored as a multi-resolution voxel space that includes a plurality of semantic layers. In some instances, individual semantic layers may include multiple voxel grids having differing resolutions. Multiple multi-resolution voxel spaces may be merged to generate combined scenes based on detected voxel covariances at one or more resolutions.

Abstract: Techniques for determining parking spaces and/or parking trajectories for a vehicle based on multistage filtering are discussed herein. In some examples, a vehicle may navigate to a destination within an environment. The vehicle may receive a set of parking locations and perform filtering operations to determine a subset of the parking locations. For example, the vehicle may use a list of parking characteristics (e.g., conditions) to filter the set of parking spaces. Based on determining the subset of parking spaces, the vehicle may generate a set of candidate trajectories to the subset of parking spaces. The vehicle may determine a subset of the candidate trajectories based on filtering the set of candidate trajectories according to various constraints. Further, the vehicle may determine a cost for the subset of candidate trajectories. Based on the costs, the vehicle may determine a candidate trajectory for the vehicle to follow.

Abstract: Techniques for generating discrete track data for object(s) in an environment are described herein. For example, the techniques may include a decoder of a variable autoencoder may generate, based on first latent variable data associated with an object and a first time, first output data representing a first discrete occupancy representation for the object at the first time. A diffusion model may generate, based on the first output data, second latent variable data associated with the object and a second time subsequent to the first time. The decoder may generate, based on the second latent variable data, second output data representing a second discrete occupancy representation for the object at the second time. A track component may generate, based on the first output data and the second output data, track data including a discrete track for the object over at least the first time and the second time.

Abstract: A side-impact crash structure may be positioned proximate an end of a vehicle to reduce impact forces imparted to an occupant during a side collision. The side-impact crash structure may include energy absorbers positioned on lateral sides of the vehicle between the passenger compartment and a longitudinal end of the vehicle. The energy absorbers may be configured to absorb impact forces over a limited ride-down distance (the distance over which the deceleration occurs) to prevent damage to both the occupants and one or more components or systems of the vehicle. The energy absorbers can be configured to deform along an oblique axis of the vehicle under a compressive force. The side-impact crash structure may include one or more load spreaders disposed between the energy absorbers that disperse the impact force to other portions of the vehicle.

Abstract: Techniques for determining attributes associated with objects represented in temporal sensor data. In some examples, the techniques may include receiving sensor data including a representation of an object in an environment. The sensor data may be generated by a temporal sensor of a vehicle and, in some instances, a trajectory of the object or the vehicle may contribute to a distortion in the representation of the object. For instance, a shape of the representation of the object may be distorted relative to an actual shape of the object. The techniques may also include determining an attribute (e.g., velocity, bounding box, etc.) associated with the object based at least in part on a difference between the representation of the object and another representation of the object (e.g., in other sensor data) or the actual shape of the object.

Abstract: Techniques and methods for securing vehicle systems. For instance, an authorization system may store data representing frequencies at which destination locations are associated pick-up locations for a fleet of autonomous vehicles. The authorization system may then receive a request for an autonomous vehicle to pick up a passenger at a first location and drop off the passenger at a second location. Based on the first location and the second location, the authorization system may determine a frequency for the request. The authorization system may then determine whether a control system for the fleet of autonomous vehicles is compromised based on whether the frequency is less than or equal to a threshold frequency. If the authorization system determines that the control system is compromised, the authorization system may perform a remedial action, such as notifying a teleoperator.

Abstract: A system and method to defog and/or defrost a sealed sensor system housing window to improve transmission of an electromagnetic wave through the housing window. The system includes at least a heater provided in a sealed sensor system housing to emit heat into the housing. The heat is then distributed via convection promoted by a moving component of the sensor system, a fan, or a combination of both. Additionally or alternatively, heat may be transferred to the housing window via conduction. A controller can be configured to activate or deactivate the heater based on a determined or predicted state.

Abstract: An autonomous delivery vehicle including locking storage containers may be used for item deliveries, rejections, returns, and/or third-party fulfillment. A delivery vehicle or robot may include a number of locking storage containers, an authorization interface, and one or more sensors to receive delivery requests, detect and authorize users, and control locker access at various delivery locations to allow users to receive delivered items, and reject or return items. The vehicle may also include a passenger compartment to transport one or more passengers. The vehicle may be reconfigurable to accommodate different combinations of lockers and/or passenger seats. An item delivery system may receive delivery requests and determine routes for delivery vehicles, including centralized delivery locations and/or direct deliveries to recipients.

Abstract: A machine-learned architecture for generating a single trajectory or multiple trajectories for controlling a vehicle may comprise an embedding model that generates an embedding of a world state indicating environment, object, and/or other states, one or more machine-learned layers that determine a predicted world state embedding using the world state embedding, and concatenating, as combined data, that predicted world state embedding to a steering angle distribution and a velocity distribution. The combined data is provided as input to a machine-learned model (that may be a single machine-learned model or may comprise two separate machine-learned models) that determines a next steering angle distribution and a next velocity distribution. These distributions may be used as part of generating one or more trajectories for controlling the vehicle. The architecture may be iteratively used to create a series of distributions that are used to create one or more trajectories.

Abstract: Techniques for protecting a sensor from an obstruction such as liquid are discussed herein. For instance, a sensor cover assembly can include a cover to protect a sensor from precipitation, contaminants, or other obstructions. The cover of the sensor cover assembly can be configured to direct an obstruction (e.g., a liquid) away from a lens of the sensor (e.g., a ring lens of a lidar sensor) to reduce obstructions to a field of view of the sensor. The cover can include one or more channels (e.g., a trough, a downspout, a gutter, etc.) or shapes (e.g., a curved surface, a sloped surface, a convex surface, etc.) to divert the obstruction away from the lens of the sensor.

Abstract: This disclosure describes techniques for detecting multipath radar returns and modifying radar data. A vehicle may use radar devices to receive radar data while traversing within an environment. The vehicle may process the radar data using a virtual array based on an arraignment of the antennae within an aperture of the radar device. Using the virtual array, the vehicle may determine an elevated noise level that may be indicative of a multipath radar return. Based on the elevated noise level, the vehicle may determine a second virtual array associated with multipath radar returns, and may process the radar data using the second virtual array. Based on determining that the noise level associated with the second virtual array is lower than the initial noise level, the vehicle may determine that the radar data includes a multipath radar return, and may modify the radar data to correct or mitigate the error caused by the multipath return.

Abstract: Techniques are described for uniquely identifying a quadrant of a symmetrical vehicle, such as a symmetrical autonomous vehicle. The techniques include using service indicator lights to convey information regarding the status and operational mode of the vehicle to an observer and/or service technician. The service indicator lights may uniquely identify quadrants of the vehicle for orienting the vehicle with a representation of the vehicle to guide a technician to a particular location and component of the vehicle located within the identified quadrant.

Abstract: Techniques for determining relevant objects in an environment around a vehicle are discussed herein. Sensor data can be captured and input to a model to determine predicted object trajectories and associated probabilities. For a candidate trajectory for the vehicle to follow, regions of overlap between the vehicle trajectory and the predicted object trajectories can be determined. An interaction graph can be generated representing interactions between the candidate trajectory and the predicted object trajectories. The interaction graph can be sampled based on the probabilities and/or other characteristics. Samples from the interaction graph can be used to determine relevant objects in a driving scenario for a vehicle traversing an environment. Based on the sample, the vehicle may generate a relevancy score associated with the sample. The vehicle may determine that a number of the most relevant sample scenarios may be sent to a prediction component and control of the vehicle.

Abstract: Techniques for determining detection boxes representing objects in an environment using pixel clustering 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 pixels that are associated with a large object. A vehicle computing system can generate a detection box for each pixel of the plurality of pixels and can determine at least one averaged detection box representing the large object. In an example, the vehicle computing system can control the vehicle based in part on the at least one averaged detection boxes.

Abstract: Techniques for determining a parking trajectory for an autonomous vehicle. In some examples, a vehicle may navigate to a destination within an environment. The vehicle may receive parking locations proximate the destination, and determine a cost for navigating to each parking location, or multiple locations within a parking space. Each of multiple trajectories may include desired state(s) that represent the desired state information of the vehicle at a specific locations along the trajectory. The vehicle may determine a cost for the trajectory by comparing the predicted pose of the vehicle at the parking state to the target pose of the parking location. The vehicle may follow the trajectory based on the cost.

Abstract: Techniques are discussed for predicting an occupancy of visible region of an environment. For instance, a vehicle may generate sensor data representing an environment. The vehicle may then analyze the sensor data to determine an occluded region of the environment a visible region of the environment. Additionally, the vehicle may determine at least one prediction probability associated with occupancy of the visible region over a future period of time. In some instances, the vehicle determines the at least one prediction probability by inputting data representing at least the occluded region and the visible region into a machine learned model and receiving the at least one prediction probability from the machine learned model. Using the at least one prediction probability, the vehicle may then determine and perform one or more actions.

Abstract: Techniques for determining parking spaces and/or parking trajectories for a vehicle based on multistage filtering are discussed herein. In some examples, a vehicle may navigate to a destination within an environment. The vehicle may receive a set of parking locations and perform filtering operations to determine a subset of the parking locations. For example, the vehicle may use a list of parking characteristics (e.g., conditions) to filter the set of parking spaces. Based on determining the subset of parking spaces, the vehicle may generate a set of candidate trajectories to the subset of parking spaces. The vehicle may determine a subset of the candidate trajectories based on filtering the set of candidate trajectories according to various constraints. Further, the vehicle may determine a cost for the subset of candidate trajectories. Based on the costs, the vehicle may determine a candidate trajectory for the vehicle to follow.