ZENSEACT AB

The present invention is related to a computer-implemented method for generating synthetic traffic sign data for training a traffic sign detection model. The method includes: extracting, from a sample image depicting a surrounding environment of a vehicle including a detected traffic sign, a face of the detected traffic sign, using information indicative of an orientation of the detected traffic sign; forming a 3D model of the traffic sign having the extracted face of the detected traffic sign; and generating synthetic traffic sign data by placing the 3D model of the traffic sign in one or more background images, wherein the 3D model is placed in the one or more background images based on a distribution of a placement of real-world traffic signs in a dataset of real-world images.

A method and a computing device for expanding a scenario database is disclosed. The scenario database includes a plurality of existing scenario samples, each scenario sample including sensor data depicting a surrounding environment of a vehicle over a period of time, wherein each scenario sample is associated with a scenario embedding representing the scenario sample in a multi-dimensional space. The method includes obtaining an indication of an empty volume in the multi-dimensional space, the empty volume is a volume not covered by the plurality of existing scenario samples of the scenario database; transmitting a data collection request in view of the empty volume to one or more vehicles; receiving, from a vehicle of a fleet of vehicles, data indicative of a recorded scenario matching the data collection request; and storing the received data indicative of the recorded scenario to the plurality of existing scenario samples of the scenario database.

A method for generating synthetic sensor data using a scenario database is disclosed. The database includes a plurality of scenario samples, each comprising sensor data depicting a vehicle's surrounding environment over time. Each scenario sample has an associated scenario embedding in a multi-dimensional space and a corresponding transformation volume indicating possible transformed scenarios derived from that sample. The method includes obtaining a request specifying a query scenario, the query scenario having a query embedding in the multi-dimensional space. The method further includes identifying, within the scenario database, at least one scenario sample whose transformation volume contains the query embedding. In response to identifying such a scenario sample, the method generates synthetic sensor data for the query scenario by transforming the sensor data of the identified scenario sample so that the synthetic sensor data has an associated scenario embedding within a threshold distance of the query embedding.

A method for generating predictive output for an ADS of a vehicle is disclosed. The method includes, by a first hardware platform, encoding a first sensor dataset from a first cluster of sensors using first encoder networks and fusing the encoded data using a first sensor fusion network to form first fused sensor data features. The method further includes, by a second hardware platform, encoding a second sensor dataset from a second cluster of sensors using second encoder networks and fusing the encoded data using a second sensor fusion network to form second fused sensor data features. The first and second hardware platforms are separate, and each sensor in the first cluster differs from those in the second cluster. Thereafter, the first and second fused feature sets are fused using a third sensor fusion network to form third fused features, and generate a predictive output using decoder networks.

A method is provided for the efficient training of a neural network for an Automated Driving System (ADS) of a vehicle, utilizing a Central Processing Unit (CPU) to handle input/output (I/O) operations for reading and writing sensor data from a disk, and a Graphics Processing Unit (GPU) for performing the network training. The method includes monitoring a GPU utilization value, which reflects the current or queued workload of the GPU, to ensure continuous, high-throughput training. If the monitored GPU utilization value is above a threshold, then the method performs the steps of reading the sensor data from the disk using the CPU, transferring it to the GPU, and training the neural network accordingly. If value is below the threshold, then the method performs the steps of generating synthetic sensor data by using the GPU, and training the neural network based on the synthetic sensor data by the GPU.

Computer-implemented methods and related aspects for providing trajectories for execution by a vehicle are disclosed. The computer-implemented method includes obtaining a free-space area in a surrounding environment of the vehicle that is formed based on sensor data output from one or more vehicle-mounted sensors. The obtained free-space area is defined by an edge having one or more edge property classes that are defined by corresponding object classes of objects that define segments of the edge of the obtained free-space area. The computer-implemented method further includes updating the obtained free-space area by modifying a size of the obtained free-space area based on the one or more edge property classes, and determining or approving a vehicle speed for a candidate path within the updated free-space area to be executed by the vehicle. The vehicle speed is determined in dependence of the determined modification of the size of obtained free-space area.

The present disclosure relates to methods, systems, a vehicle and a computer-readable storage medium and a computer program product. The method includes obtaining a baseline risk model generated based on one or more risk-associated parameters. The method further includes obtaining sensor data from a sensor system of a vehicle. The method further includes estimating one or more risk values associated with the one or more risk-associated parameters based on the obtained sensor data and generating, based on the one or more estimated risk values, an adopted risk model. Further, the method includes provisioning the generated adopted risk model for determining an acceptable risk model for forming a driving policy of an ADS.

A method for closed-loop evaluation of path planning modules for a vehicle equipped with an Automated Driving System (ADS) is disclosed. The method includes obtaining a candidate path from each of a plurality of path planning modules of the ADS. The method further includes determining a fulfilment of convergence criteria by the obtained candidate paths by comparing the obtained candidate paths with each other and determining a level of convergence between the candidate paths. If the convergence criteria is fulfilled, one of the obtained candidate paths is selected and the vehicle is controlled so to execute the selected candidate path. If the convergence criteria is not fulfilled, determining an exposure need of each path planning module in view of a predicted scene or scenario in the surrounding environment of the vehicle that the vehicle is expected to be exposed to while executing any one of the obtained candidate paths.

A computer-implemented method for classifying a traffic sign in an image, a computing device and vehicle thereof is disclosed. The method includes obtaining the image depicting at least a portion of a surrounding environment of the vehicle; identifying a region in the image corresponding to a traffic sign, by processing the image through a first machine learning model configured to output detections of traffic signs in input images; extracting a crop corresponding to the identified region, wherein the crop has a native resolution based on a size of the identified region in relation to the obtained image; and determining classification data of the traffic sign by processing the crop, at the native resolution, through a second machine learning model, wherein the second machine learning model is an attention-based neural network, trained to process input images of traffic signs of varying resolution and to generate corresponding classification data.

A method for generating prediction output for an Automated Driving System (ADS) of a vehicle is disclosed. The method includes obtaining, by one or more processors, a first sensor dataset from a first sensor and a second sensor dataset from a different sensor, each including information about a portion of a surrounding environment of the vehicle. The method further includes processing the first sensor dataset using a first ensemble of two or more encoder networks, each trained to output a first set of encoded features, and processing the second dataset using a second ensemble of two or more encoder networks, each trained to output a second set of encoded features. Then, one or more sets of the encoded features from the first and second datasets are fused using a fusion algorithm to output fused encoded features. A decoder network then generates a prediction output based on the fused encoded features.