Abstract: A method of determining a navigation trajectory for an autonomous ground vehicle (AGV) is disclosed. The method may include receiving first Region of Interest (ROI) data associated with an upcoming trajectory path, and receiving predicted attributes associated with a future navigation trajectory for the upcoming trajectory path. The predicted attributes are derived based on map for the upcoming trajectory path. The method may further include modifying the predicted attributes based on environmental attributes extracted from first ROI data to generate modified attributes, and dynamically receiving a second ROI data associated with the upcoming trajectory path upon reaching the upcoming trajectory path. The method may further include predicting dynamic attributes associated with an imminent navigation trajectory for the upcoming trajectory path based on the second ROI data, and refining the modified attributes based on the one or more dynamic attributes to generate a final navigation trajectory.

Abstract: A method and system for determining lane change feasibility for autonomous vehicles is disclosed. The method includes the steps of tracking, in each frame, at least one neighboring vehicle from a plurality of neighboring vehicles in a current lane being used by the AV and in a plurality of adjacent lanes. The method further includes determining, in each frame, a set of kinematic parameters associated with each of the at least one neighboring vehicle, and assigning an occupancy state from a plurality of occupancy states to each of a plurality of voxels capturing a spatial information for each of the plurality of neighboring vehicles. The method may further includes determining an effective occupancy probability for each of the plurality of adjacent lanes, and determining feasibility of lane change for the AV to at least one adjacent lane from the plurality of adjacent lanes.

Abstract: The present disclosure relates to a method of stitching images captured by a vehicle. A first image and a second image are received. The first image and the second image are segmented based on characteristics of pixels. Groups of pixels having similar characteristics are identified to form clusters in a predetermined portion of overlap of the first image and the second image. A confidence score is generated for the first image and the second image. A difference in the confidence score is computed. At least one of, the first image capturing unit and the second image capturing unit is aligned to capture at least one of, a first aligned image and a second aligned image based on the difference in the confidence score. The first aligned image and the second aligned image are stitched.

Abstract: A method and system for determining lane change feasibility for autonomous vehicles is disclosed. The method includes the steps of tracking, in each frame, at least one neighboring vehicle from a plurality of neighboring vehicles in a current lane being used by the AV and in a plurality of adjacent lanes. The method further includes determining, in each frame, a set of kinematic parameters associated with each of the at least one neighboring vehicle, and assigning an occupancy state from a plurality of occupancy states to each of a plurality of voxels capturing a spatial information for each of the plurality of neighboring vehicles. The method may further includes determining an effective occupancy probability for each of the plurality of adjacent lanes, and determining feasibility of lane change for the AV to at least one adjacent lane from the plurality of adjacent lanes.

Abstract: The present disclosure relates to a method of stitching images captured by a vehicle. A first image and a second image are received. The first image and the second image are segmented based on characteristics of pixels. Groups of pixels having similar characteristics are identified to form clusters in a predetermined portion of overlap of the first image and the second image. A confidence score is generated for the first image and the second image. A difference in the confidence score is computed. At least one of, the first image capturing unit and the second image capturing unit is aligned to capture at least one of, a first aligned image and a second aligned image based on the difference in the confidence score. The first aligned image and the second aligned image are stitched.

Abstract: A method of determining a navigation trajectory for an autonomous ground vehicle (AGV) is disclosed. The method may include receiving first Region of Interest (ROI) data associated with an upcoming trajectory path, and receiving predicted attributes associated with a future navigation trajectory for the upcoming trajectory path. The predicted attributes are derived based on map for the upcoming trajectory path. The method may further include modifying the predicted attributes based on environmental attributes extracted from first ROI data to generate modified attributes, and dynamically receiving a second ROI data associated with the upcoming trajectory path upon reaching the upcoming trajectory path. The method may further include predicting dynamic attributes associated with an imminent navigation trajectory for the upcoming trajectory path based on the second ROI data, and refining the modified attributes based on the one or more dynamic attributes to generate a final navigation trajectory.

Abstract: This disclosure relates generally to generating synthetic data, and more particularly to method and system for synthesizing three-dimensional (3D) data using generative adversarial networks (GANs). The method may include clustering initial 3D data to identify one or more regions of interest (ROIs), generating an input specific noise data based on the one or more ROIs by an iterative process using Gaussian mixture model, and iteratively synthesizing the 3D data based on the one or more ROIs and the input specific noise data using generative adversarial networks (GANs) to generate final synthesized 3D data. The initial 3D data may represent a given scenario, while the final synthesized 3D data may represent a number of possible scenarios and are affine transforms of the initial 3D data.

Abstract: The present disclosure discloses method and a road region determination system for determining drivable road regions for safe navigation of an autonomous vehicle. The road region determination system receives real-time images of road in which autonomous vehicle is travelling, from image sensors. Each of real-time images of road is segmented into polygon regions and trend lines to obtain plurality of features using pre-trained road segmentation model. An orientation of the road is identified in real-time images as one of linear orientation and non-linear orientation based on slope measured between successive intermediate points distributed evenly on trend lines. The road region determination system manages redistribution of intermediate points on trend lines based on orientation of the road. Thereafter, paired points from intermediate points redistributed on the trend lines is identified, where paired points are connected using a horizontal line to determine drivable road regions for autonomous vehicle.

Abstract: This disclosure relates generally to generating synthetic data, and more particularly to method and system for synthesizing three-dimensional (3D) data using generative adversarial networks (GANs). The method may include clustering initial 3D data to identify one or more regions of interest (ROIs), generating an input specific noise data based on the one or more ROIs by an iterative process using Gaussian mixture model, and iteratively synthesizing the 3D data based on the one or more ROIs and the input specific noise data using generative adversarial networks (GANs) to generate final synthesized 3D data. The initial 3D data may represent a given scenario, while the final synthesized 3D data may represent a number of possible scenarios and are affine transforms of the initial 3D data.