Abstract: A sensor simulation system may generate sensor data for use in simulations by rendering two-dimensional views of a three-dimensional simulated environment. In various examples, the sensor simulation system uses sensor dependency data to determine specific views to be re-rendered at different times during the simulation. The sensor simulation system also may generate unified views with multi-sensor data at each region (e.g., pixel) of the two-dimensional view for consumption by different sensor types. A hybrid technique may be used in some implementations in which rasterization is used to generate a view, after which ray tracing is used to align the view with a particular sensor. Spatial and temporal upsampling techniques also may be used, including depth-aware and velocity-aware analyses for simulated objects, to improve view resolution and reduce the frequency of re-rendering views.

Abstract: Techniques for a perception system of a vehicle that can detect and track objects in an environment are described herein. The perception system may include a machine-learned model that includes one or more different portions, such as different components, subprocesses, or the like. In some instances, the techniques may include training the machine-learned model end-to-end such that outputs of a first portion of the machine-learned model are tailored for use as inputs to another portion of the machine-learned model. Additionally, or alternatively, the perception system described herein may utilize temporal data to track objects in the environment of the vehicle and associate tracking data with specific objects in the environment detected by the machine-learned model. That is, the architecture of the machine-learned model may include both a detection portion and a tracking portion in the same loop.

Abstract: Techniques for detecting attributes and/or gestures associated with pedestrians in an environment are described herein. The techniques may include receiving sensor data associated with a pedestrian in an environment of a vehicle and inputting the sensor data into a machine-learned model that is configured to determine a gesture and/or an attribute of the pedestrian. Based on the input data, an output may be received from the machine-learned model that indicates the gesture and/or the attribute of the pedestrian and the vehicle may be controlled based at least in part on the gesture and/or the attribute of the pedestrian. The techniques may also include training the machine-learned model to detect the attribute and/or the gesture of the pedestrian.

Abstract: Techniques to predict object behavior in an environment are discussed herein. For example, such techniques may include determining a trajectory of the object, determining an intent of the trajectory, and sending the trajectory and the intent to a vehicle computing system to control an autonomous vehicle. The vehicle computing system may implement a machine learned model to process data such as sensor data and map data. The machine learned model can associate different intentions of an object in an environment with different trajectories. A vehicle, such as an autonomous vehicle, can be controlled to traverse an environment based on object's intentions and trajectories.

Abstract: A vehicle mounted sensor assembly includes a frame and a cover connected to the base. When assembled, the frame and the cover define a volume that is divided between a first cavity and a second cavity. A sensor is disposed in the first cavity and one or more antennas are disposed in the second cavity.

Abstract: Techniques for determining a probability of a false negative associated with a location of an environment are discussed herein. Data from a sensor, such as a radar sensor, can be received that includes point cloud data, which includes first and second data points. The first data point has a first attribute and the second data point has a second attribute. A difference between the first and second attributes is determined such that a frequency distribution may be determined. The frequency distribution may then be used to determine a distribution function, which allows for the determination of a resolution function that is associated with the sensor. The resolution function may then be used to determine a probability of a false negative at a location in an environment. The probability can be used to control a vehicle in a safe and reliable manner.

Abstract: A machine-learning architecture may be trained to determine point cloud data associated with different types of sensors with an object detected in an image and/or generate a three-dimensional region of interest (ROI) associated with the object. In some examples, the point cloud data may be associated with sensors such as, for example, a lidar device, radar device, etc.

Abstract: Techniques for selectively offloading data that is computed by a first processing unit during training of an artificial neural network onto memory associated with a second processing unit and transferring the data back to the first processing unit when the data is needed for further processing are described herein. For example, the first processing unit may compute activations for operations associated with forward propagation. During the forward propagation, one or more of the activations may be transferred to a second processing unit for storage. Then, during backpropagation for the artificial neural network, the activations may be transferred back to the first processing unit as needed to compute gradients.

Abstract: The described techniques relate to modifying a trajectory of a vehicle, such as an autonomous vehicle, based on a latency associated with one or more systems of the vehicle. In examples, a planning system of the vehicle may predict a future latency (e.g., based on an interval between receipt of sensor data and/or object predictions), and use the future latency to determine a time at which to predict object behavior. Additionally, in some cases, the described techniques may include associating a predetermined acceleration with a predicted future location of the object to create a safety distance around the object, where the predetermined acceleration may be based on a maximum expected acceleration of the object. The safety distance may account for the object potentially accelerating in one or more directions at the future time.

Abstract: An autonomous vehicle safety system may activate to prevent collisions by detecting that a planned trajectory may result in a collision. If the safety system is overly sensitive, it may cause false positive activations, and if the system isn't sensitive enough the collision avoidance system may not activate and prevent a collision, which is unacceptable. It may be impossible or prohibitively difficult to detect false positive activations of a safety system and it is unacceptable to risk a false negative, so tuning the safety system is notoriously difficult. Tuning the safety system may include detecting near-miss events using surrogate metrics, and tuning the safety system to increase or decrease a rate of near-miss events as a stand-in for false positives.

Abstract: Techniques are disclosed for implementing a neural network that outputs embeddings. Furthermore, techniques are disclosed for using sensor data to train a neural network to learn such embeddings. In some examples, the neural network may be trained to learn embeddings for instance segmentation of an object based on an embedding for a bounding box associated with the object being trained to match pixel embeddings for pixels associated with the object. The embeddings may be used for object identification, object matching, object classification, and/or object tracking in various examples.

Abstract: Techniques and methods for performing collision monitoring using error models. For instance, a vehicle may generate sensor data using one or more sensors. The vehicle may then analyze the sensor data using systems in order to determine parameters associated with the vehicle and parameters associated with another object. Additionally, the vehicle may process the parameters associated with the vehicle using error models associated with the systems in order to determine a distribution of estimated locations associated with the vehicle. The vehicle may also process the parameters associated with the object using the error models in order to determine a distribution of estimated locations associated with the object. Using the distributions of estimated locations, the vehicle may determine the probability of collision between the vehicle and the object.

Abstract: A sealing system for a vehicle is configured to separate a vehicle cabin between a dry area and a wet area and to channel a spilled fluid or waste material inside the wet area to a collection area for cleaning. The sealing system may include one or more sealing members that can be extrusion molded to fill one or more gaps between two or more vehicle components.

Abstract: Techniques are described for detecting whether a lane of a roadway is open or closed. Detecting a lane as being closed may include detecting an object in or near the lane, which may comprise determining a size, location, and/or classification associated with the object, and dilating the size associated with the object. The lane may be indicated as being closed if a distance between a dilated object detection and another object detection, dilated object detection, or lane extent is less than a threshold distance. The techniques may additionally or alternatively comprise determining an alternative lane shape based at least in part on one or more object detections and/or determining that one or more lanes are closed and/or uploading a lane closure and/or alternative lane shape to a central database for retrieval by/dissemination to other computing devices.

Abstract: A three dimensional bounding box is determined from a two dimensional image. A two dimensional bounding box is calculated based on a detected object within the image. A three dimensional bounding box is parameterized as having a yaw angle, dimensions, and a position. The yaw angle is defined as the angle between a ray passing through a center of the two dimensional bounding box and an orientation of the three dimensional bounding box. The yaw angle and dimensions are determined by passing the portion of the image within the two dimensional bounding box through a trained convolutional neural network. The three dimensional bounding box is then positioned such that the projection of the three dimensional bounding box into the image aligns with the two dimensional bounding box previously detected. Characteristics of the three dimensional bounding box are then communicated to an autonomous system for collision and obstacle avoidance.

Abstract: Vehicles may be composed of a relatively few number of “modules” that are assembled together during a final assembly process. An example vehicle may include a body module, a first drive module coupled to a first end of the body module, and a second drive module coupled to a second end of the body module. One or both of the drive modules may include a pair of wheels, a battery, an electric drive motor, and/or a heating ventilation and air conditioning (HVAC) system. One or both of the drive modules may also include a crash structure to absorb impacts. If a component of a drive module fails or is damaged, the drive module can be quickly and easily replaced with a new drive module, minimizing vehicle down time.

Abstract: A seat for a vehicle may include a seatback including one or more inflatable bladders that may be configured to deploy in response to a triggering signal indicative of a change of velocity, collision, or predicted collision event. As at least a portion of the back of an occupant of the seat pushes against a front surface of the seatback due to the event, the one or more inflatable bladders compress at least partially the comfortable foam or comfort material in the seatback and expedite engagement of the energy absorbing material.