This application is directed to augmenting training data used for vehicle driving modelling. A computer system obtains a first image of a road and identifies a drivable area of the road within the first image. The computer system obtains an image of an object and generates a second image from the first image by overlaying the image of the object over the drivable area. The second image is added to a corpus of training images to be used by a machine learning system to generate a model for facilitating driving of a vehicle (e.g., at least partial autonomously). In some embodiments, the computer system applies machine learning to train a model using the corpus of training images and distributes the model to one or more vehicles. In use, the model processes road images captured by the one or more vehicles to facilitate vehicle driving.

In various examples, one or more deep neural networks (DNNs) are executed to regress on control points of a curve, and the control points may be used to perform a curve fitting operation—e.g., Bezier curve fitting—to identify landmark locations and geometries in an environment. The outputs of the DNN(s) may thus indicate the two-dimensional (2D) image-space and/or three-dimensional (3D) world-space control point locations, and post-processing techniques—such as clustering and temporal smoothing—may be executed to determine landmark locations and poses with precision and in real-time. As a result, reconstructed curves corresponding to the landmarks—e.g., lane line, road boundary line, crosswalk, pole, text, etc.—may be used by a vehicle to perform one or more operations for navigating an environment.

Aspects of the disclosure relate to classifying the status of objects. For examples, one or more computing devices detect an object from an image of a vehicle's environment. The object is associated with a location. The one or more computing devices receive data corresponding to the surfaces of objects in the vehicle's environment and identifying data within a region around the location of the object. The one or more computing devices also determine whether the data within the region corresponds to a planar surface extending away from an edge of the object. Based on this determination, the one or more computing devices classify the status of the object.

An example method includes storing marking data to specify at least one selected marking to apply at a target location along a vehicle path of travel, the marking data including a machine-readable description and a marking reference coordinate frame for the selected marking. The method also includes generating task plan data to apply the selected marking based on the marking data and at least one parameter of an application tool. The method also includes determining a location and orientation of the application tool with respect to the vehicle path of travel based on location data representing a current location of a vehicle carrying the application tool. The method also includes computing a joint-space trajectory to enable the application tool to apply the selected marking at the target location based on the task plan data and the determined location of the application tool.

Described is a system for detecting backdoor attacks in deep convolutional neural networks (CNNs). The system compiles specifications of a pretrained CNN into an executable model, resulting in a compiled model. A set of Universal Litmus Patterns (ULPs) are fed through the compiled model, resulting in a set of model outputs. The set of model outputs are classified and used to determine presence of a backdoor attack in the pretrained CNN. The system performs a response based on the presence of the backdoor attack.

A navigation system for a host vehicle may include at least one processor comprising circuitry and a memory. The memory may include instructions that when executed by the circuitry cause the at least one processor to receive from an image capture device associated with the host vehicle a captured image representative of an environment of the host vehicle, to identify a first segment of the captured image associated with a traffic light, to provide the first segment of the captured image to a first trained network, the first trained network being configured to generate a first output indicative of a state of the traffic light, to identify a second segment of the captured image that includes contextual information associated with the traffic light, to provide the second segment to a second trained network, the second trained network being configured to generate a second output indicative of a proposed navigational action for the host vehicle relative to the traffic light, to determine, based on both the first output from the first trained network and the second output from the second trained network a planned navigational action for the host vehicle and to cause the host vehicle to take the planned navigational action.

This disclosure provides a method and apparatus for recognizing a parking space.

An aspect of the present disclosure provides a method, performed by an apparatus of a host vehicle, for recognizing a parking space, the method including: recognizing another vehicle and a pillar based on data acquired from a light detection and ranging (lidar) device; recognizing a parking slot marking based on an image captured by a camera; and generating a map of an indoor parking lot based on information on the another vehicle, the pillar, and the parking slot marking.

The present disclosure relates to systems, methods, and non-transitory computer readable media for identifying traffic control features based on telemetry patterns within digital image representations of vehicle telemetry information. The disclosed systems can generate a digital image representation based on collected telemetry information to represent the frequency of different speed-location combinations for transportation vehicles passing through a traffic area. The disclosed systems can also apply a convolutional neural network to analyze the digital image representation and generate a predicted classification of a type of traffic control feature that corresponds to the digital image representation of vehicle telemetry information. The disclosed systems further train the convolutional neural network to determine traffic control features based on training data.

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.

The present disclosure relates generally to identification of drivable surfaces in connection with autonomously performing various tasks at industrial work sites and, more particularly, to techniques for distinguishing drivable surfaces from non-drivable surfaces based on sensor data. A framework for the identification of drivable surfaces is provided for an autonomous machine to facilitate it to autonomously detect the presence of a drivable surface and to estimate, based on sensor data, attributes of the drivable surface such as road condition, road curvature, degree of inclination or declination, and the like. In certain embodiments, at least one camera image is processed to extract a set features from which surfaces and objects in a physical environment are identified, and to generate additional images for further processing. The additional images are combined with a 3D representation, derived from LIDAR or radar data, to generate an output representation indicating a drivable surface.