Abstract: A method for image style transfer using a Semantic Preserved Generative Adversarial Network (SPGAN) includes: receiving a source image; inputting the source image into the SPGAN; extracting a source-semantic feature data from the source image; generating, by the first decoder, a first synthetic image including the source semantic content of the source image in a target style of a target image using the source-semantic feature data extracted by the first encoder of the first generator network, wherein the first synthetic image includes first-synthetic feature data; determining a first encoder loss using the source-semantic feature data and the first-synthetic feature data; discriminating the first synthetic image against the target image to determine a GAN loss; determining a total loss as a function of the first encoder loss and the first GAN loss; and training the first generator network and the first discriminator network.

Abstract: Systems and methods to identify an attention region in sensor-based detection involve obtaining a detection result that indicates one or more detection areas where one or more objects of interest are detected. The detection result is based on using a first detection algorithm. The method includes obtaining a reference detection result that indicates one or more reference detection areas where one or more objects of interest are detected. The reference detection result is based on using a second detection algorithm. The method also includes identifying the attention region as one of the one or more reference detection areas without a corresponding one or more detection areas. The first detection algorithm is used to perform detection in the attention region.

Abstract: A risk maneuver assessment system and method to generate a perception of an environment of a vehicle and a behavior decision making model for the vehicle; a sensor system configured to provide the sensor input in the environment for filtering target objects; one or more modules configured to map and track target objects to make a candidate detection from multiple candidate detections of a true candidate detection as the tracked target object; apply a Markov Random Field (MRF) algorithm for recognizing a current situation of the vehicle and predict a risk of executing a planned vehicle maneuver at the true detection of the dynamically tracked target; apply mapping functions to sensed data of the environment for configuring a machine learning model of decision making behavior of the vehicle; and apply adaptive threshold to cells of an occupancy grid for representing an area of tracking of objects within the vehicle environment.

Abstract: Systems and methods to generate an adversarial attack on a black box object detection algorithm of a sensor involve obtaining an initial training data set from the black box object detection algorithm. The black box object detection algorithm performs object detection on initial input data to provide black box object detection algorithm output that provides the initial training data set. A substitute model is trained with the initial training data set such that output from the substitute model replicates the black box object detection algorithm output that makes up the initial training data set. Details of operation of the black box object detection algorithm are unknown and details of operation of the substitute model are known. The substitute model is used to perform the adversarial attack. The adversarial attack refers to identifying adversarial input data for which the black box object detection algorithm will fail to perform accurate detection.

Abstract: An adaptive sensor control system for a vehicle includes a controller and a steerable sensor system. The controller generates a perception of the vehicle's environment, including providing at least one perception datum and an associated uncertainty factor for different areas within the perception of the environment of the vehicle. The controller also determines one or more relevance factor for the different areas within the perception of the environment. Furthermore, the controller generates control commands for steering the sensor system toward a physical space in the environment as a function of the uncertainty factor and one or more relevance factors. Accordingly, the sensor system obtains updated sensor input for the physical space to update the perception datum and the associated uncertainty factor for the physical space.

Abstract: A system and method to resolve ambiguity in a radar system involve detecting one or more objects with the radar system. The detecting includes obtaining range, azimuth, and an ambiguous range rate of a first object of the one or more objects. A plurality of Kalman filters are generated with state variables that include parameters based on the range, the azimuth, and the ambiguous range rate. Each of the plurality of Kalman filters provides a different estimate for an unambiguous range rate. The method also includes updating the plurality of Kalman filters using additional detections by the radar system, selecting a selected Kalman filter from among the plurality of Kalman filters that exhibits a highest probability mass among a plurality of probability mass corresponding with and derived from the plurality of Kalman filters, and determining the unambiguous range rate of the object using the selected Kalman filter.

Abstract: Technical solutions are described for controlling an automated driving system of a vehicle. An example method includes computing a complexity metric of an upcoming region along a route that the vehicle is traveling along. The method further includes, in response to the complexity metric being below a predetermined low-complexity threshold, determining a trajectory for the vehicle to travel in the upcoming region using a computing system of the vehicle. Further, the method includes in response to the complexity metric being above a predetermined high-complexity threshold, instructing an external computing system to determine the trajectory for the vehicle to travel in the upcoming region. If the trajectory cannot be determined by the external computing system a minimal risk condition maneuver of the vehicle is performed.

Abstract: Technical solutions are described for vehicle collision prevention for a vehicle when the vehicle is in a parked condition. An example method includes performing a stationary safety monitoring when the vehicle is in parked condition. The stationary safety monitoring includes detecting presence of a moving object within a predetermined region from the vehicle. Further, the method includes, in response to detecting the moving object in the predetermined region initiating a notification for the moving object to prevent collision with the vehicle.

Abstract: A system and method for generating a range image using sparse depth data is disclosed. The method includes receiving, by a controller, image data of a scene. The image data includes a first set of pixels. The method also includes receiving, by the controller, a sparse depth data of the scene. The sparse depth data includes a second set of pixels, and the number of the second set of pixels is less than the number of first set of pixels. The method also includes combining the image data and the sparse depth data into a combined data. The method also includes generating a range image using the combined data.

Abstract: Systems and methods are provided for controlling an autonomous vehicle (AV). A map generator module processes sensor data to generate a world representation of a particular driving scenario (PDS). A scene understanding module (SUM) processes navigation route data, position information and a feature map to define an autonomous driving task (ADT), and decomposes the ADT into a sequence of sub-tasks. The SUM selects a particular combination of sensorimotor primitive modules (SPMs) to be enabled and executed for the PDS. Each one of the SPMs addresses a sub-task in the sequence. A primitive processor module executes the particular combination of the SPMs such that each generates a vehicle trajectory and speed (VTS) profile. A selected one of the VTS profiles is then processed to generate the control signals, which are then processed at a low-level controller to generate commands that control one or more of actuators of the AV.

Abstract: An autonomic vehicle control system includes a perception module of a spatial monitoring system that is disposed to monitor a spatial environment proximal to the autonomous vehicle. A method for evaluating vehicle dynamics operation includes determining a desired trajectory for the autonomous vehicle, wherein the desired trajectory includes desired vehicle positions including an x-position, a y-position and a heading. Vehicle control commands are determined based upon the desired trajectory, and include a commanded steering angle, an acceleration command and a braking command. Actual vehicle states responsive to the vehicle control commands are determined. An estimated trajectory is determined based upon the actual vehicle states, and a trajectory error is determined based upon a difference between the desired trajectory and the estimated trajectory. The trajectory error is monitored over a time horizon, and a first state of health (SOH) is determined based upon the trajectory error over the time horizon.

Abstract: A system and method of operating a vehicle. The system includes a two-dimensional imager, a three-dimensional imager, and at least one processor. The two-dimensional imager obtains a two-dimensional image of an environment surrounding the vehicle, wherein the environment includes an object. The three-dimensional imager obtains a three-dimensional (3D) point cloud of the environment. The at least one processor identifies the object from the 2D image and assigns the identification of the object to a selected point of the 3D point cloud.

Abstract: A method and system including a central processing unit (CPU), an accelerator, a communication bus and a system memory device for dynamically processing an image file are described. The accelerator includes a local memory buffer, a data transfer scheduler, and a plurality of processing engines. The data transfer scheduler is arranged to manage data transfer between the system memory device and the local memory buffer, wherein the data transfer includes data associated with the image file. The local memory buffer is configured as a circular line buffer, and the data transfer scheduler includes a ping-pong buffer for transferring output data from the one of the processing engines to the system memory device. The local memory buffer is configured to execute cross-layer usage of data associated with the image file.

Abstract: Systems and methods for depth estimation of images from a mono-camera by use of radar data by: receiving, a plurality of input 2-D images from the mono-camera; generating, by the processing unit, an estimated depth image by supervised training of an image estimation model; generating, by the processing unit, a synthetic image from a first input image and a second input image from the mono-camera by applying an estimated transform pose; comparing, by the processing unit, an estimated three-dimensional (3-D) point cloud to radar data by applying another estimated transform pose to a 3-D point cloud wherein the 3-D point cloud is estimated from a depth image by supervised training of the image estimation model to radar distance and radar doppler measurement; correcting a depth estimation of the estimated depth image by losses derived from differences: of the synthetic image and original images; of an estimated depth image and a measured radar distance; and of an estimated doppler information and measured radar d

Abstract: An ultra-short range radar (USRR) system of a vehicle includes an object detection module configured to, based on radar signals from USRR sensors of the vehicle: identify the presence of an object that is external to the vehicle; determine a location of the object; and determine at least one of a height, a length, and a width of the object. A remedial action module is configured to, based on the location of the object and the at least one dimension of the object, at least one of: selectively actuate an actuator of the vehicle; selectively generate an audible alert via at least one speaker of the vehicle; and selectively generate a visual alert via at least one light emitting device of the vehicle.

Abstract: A signal processing system includes a central processing unit (CPU) in communication with an accelerator, and an instruction scheduler in communication with the accelerator. A first memory device including a first instruction set is configured to operate the accelerator, a second instruction set is configured to operate the CPU, and a second memory device is configured to receive a datafile. The accelerator includes a plurality of processing engines (PEs) and an instruction scheduler, the instruction set includes a plurality of operators, and the instruction scheduler is configured to implement the operators in the accelerator employing the PEs. The CPU employs the operators implemented in the accelerator to analyze the datafile to extract a feature therefrom.

Abstract: A method and system including a central processing unit (CPU), an accelerator, a communication bus and a system memory device for dynamically processing an image file are described. The accelerator includes a local memory buffer, a data transfer scheduler, and a plurality of processing engines. The data transfer scheduler is arranged to manage data transfer between the system memory device and the local memory buffer, wherein the data transfer includes data associated with the image file. The local memory buffer is configured as a circular line buffer, and the data transfer scheduler includes a ping-pong buffer for transferring output data from the one of the processing engines to the system memory device. The local memory buffer is configured to execute cross-layer usage of data associated with the image file.

Abstract: A disambiguating system for disambiguating between ambiguous detections is provided. The system includes a plurality of modules, wherein each module is configured to disambiguate between ambiguous detections by selecting, as a true detection, one candidate detection in a set of ambiguous detections and wherein each module is configured to apply a different selection technique. The system includes: one or more modules configured to select as the true detection, the candidate detection whose associated position is closer to a position indicated by other data and one or more modules configured to select as the true detection, the candidate detection with the highest probability of being true based on other sensor data.

Abstract: In one example implementation according to aspects of the present disclosure, a computer-implemented method includes capturing a plurality of images at a camera associated with a vehicle and storing image data associated with the plurality of images to a memory. The method further includes dispatching vehicle perception tasks to a plurality of processing elements of an accelerator in communication with the memory. The method further includes performing, by at least one of the plurality of processing elements, the vehicle perception tasks for the vehicle perception using a neural network, wherein performing the vehicle perception tasks comprises performing an activation bypass for values below a first threshold, and performing weight pruning of synapses and neurons of the neural network based at least in part on a second threshold. The method further includes controlling the vehicle based at least in part on a result of performing the vehicle perception tasks.

Abstract: Examples of techniques for using fixed-point quantization in deep neural networks are disclosed. In one example implementation according to aspects of the present disclosure, a computer-implemented method includes capturing a plurality of images at a camera associated with a vehicle and storing image data associated with the plurality of images to a memory. The method further includes dispatching vehicle perception tasks to a plurality of processing elements of an accelerator in communication with the memory. The method further includes performing, by at least one of the plurality of processing elements, the vehicle perception tasks for the vehicle perception using a neural network, wherein performing the vehicle perception tasks comprises quantizing a fixed-point value based on an activation input and a synapse weight. The method further includes controlling the vehicle based at least in part on a result of performing the vehicle perception tasks.