Abstract: An autonomous vehicle identifies an intersection, identifies an object in proximity to the intersection, identifies a plurality of outlets of the intersection, and, for each outlet, identifies a polyline associated with the outlet, identifies a target point along the polyline, and determines a constant curvature path from the object to the target point. The system determines a score associated with each outlet based at least in part on the constant curvature path of the outlet, generates a pruned set of outlets that includes one or more of the outlets from the plurality of outlets based on its score, and for each outlet in the pruned set, generates a reference path from the object to the target point of the outlet.

Abstract: An autonomous vehicle navigates an intersection in which occlusions block the vehicle's ability to detect moving objects. The vehicle handles this by generating a phantom obstacle behind the occlusion. The vehicle will predict the speed of the phantom obstacle and use the predicted speed to assess whether the phantom obstacle may collide with the vehicle. If a collision is a risk, the vehicle will slow or stop until it confirms that either (a) the phantom obstacle is not a real obstacle or (b) the vehicle can proceed at a speed that avoids the collision. To determine which occlusions shield real objects, the system may use a rasterized visibility grid of the area to identify occlusions that may accommodate the object.

Abstract: Systems and methods for object detection. The object detection may be used to control an autonomous vehicle. For example, the methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of the autonomous vehicle; and using, by the computing device, the LiDAR dataset and at least one image to detect an object that is in proximity to the autonomous vehicle. The object is detected by: matching points of the LiDAR dataset to pixels in the at least one image; and detecting the object in a point cloud defined by the LiDAR dataset based on the matching. The object detection may be used to facilitate at least one autonomous driving operation (e.g., autonomous driving operation comprises an object tracking operation, an object trajectory prediction operation, a vehicle trajectory determination operation, and/or a collision avoidance operation).

Abstract: Systems and methods for object detection. Object detection may be used to control autonomous vehicle(s). For example, the methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of the autonomous vehicle; and using, by the computing device. The LiDAR dataset and image(s) are used to detect an object that is in proximity to the autonomous vehicle. The object is detected by: computing a distribution of object detections that each point of the LiDAR dataset is likely to be in; creating a plurality of segments of LiDAR data points using the distribution of object detections; and detecting the object in a point cloud defined by the LiDAR dataset based on the plurality of segments of LiDAR data points. The object detection may be used to facilitate at least one autonomous driving operation.

Abstract: Systems and methods for monitoring the lane of an object in an environment of an autonomous vehicle are disclosed. The methods include receiving sensor data corresponding to the object, and assigning an instantaneous probability to each of a plurality of lanes based on the sensor data as a measure of likelihood that the object is in that lane at a current time. The methods also include generating a transition matrix for each of the plurality of lanes that encode one or more probabilities that the object transitioned to that lane from another lane in the environment or from that lane to another lane in the environment at the current time. The methods then include determining an assigned probability associated with each of the plurality of lanes based on the instantaneous probability and the transition matrix as a measure of likelihood of the object occupying that lane at the current time.

Abstract: Systems and methods for object detection. Object detection may be used to control autonomous vehicle(s). For example, the methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of the autonomous vehicle; and using, by the computing device, the LiDAR dataset and image(s) to detect an object that is in proximity to the autonomous vehicle. The object being is detected by: computing a distribution of object detections that each point of the LiDAR dataset is likely to be in; creating a plurality of segments of LiDAR data points using the distribution of object detections; merging the plurality of segments of LiDAR data points to generate merged segments; and detecting the object in a point cloud defined by the LiDAR dataset based on the merged segments. The object detection may be used by the computing device to facilitate at least one autonomous driving operation.

Abstract: Systems/methods for object detection. The methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of the autonomous vehicle; and using, by a computing device, the LiDAR dataset and at least one image to detect an object that is in proximity to the autonomous vehicle. The object is detected by: generating a pruned LiDAR dataset by reducing a total number of points contained in the LiDAR dataset; and detecting the object in a point cloud defined by the pruned LiDAR dataset. The object detection may be used by the computing device to facilitate at least one autonomous driving operation.

Abstract: Systems and methods for object detection. Object detection may be used to control autonomous vehicle(s). For example, the methods comprise: obtaining, by a computing device, a LiDAR dataset generated by a LiDAR system of autonomous vehicle; and using, by the computing device, the LiDAR dataset and image(s) to detect an object that is in proximity to the autonomous vehicle. The object is detected by performing the following operations: computing a distribution of object detections that each point of the LiDAR dataset is likely to be in; creating a plurality of segments of LiDAR data points using the distribution of object detections; merging the plurality of segments of LiDAR data points to generate merged segments; and detecting the object in a point cloud defined by the LiDAR dataset based on the merged segments. The object detection may be used by the computing device to facilitate at least one autonomous driving operation.

Abstract: A system may receive point cloud data that includes one or more data points associated with an object that was detected by sensors of an autonomous vehicle. The system may identify a subset of the point cloud data having data points that are associated with a likelihood of a pedestrian entering a scene with the object, determine a current probability value using a logistic function that is associated with the subset of the point cloud data, determine, based at least in part on the current probability value, a probability value representing a likelihood of the pedestrian actually being present for the subset of the point cloud data, determine whether the probability value exceeds a false alarm threshold value, and in response to the probability value exceeding the false alarm threshold value, assign data points of the subset an attribute value indicative of the pedestrian being present.

Abstract: An autonomous vehicle identifies an intersection, identifies an object in proximity to the intersection, identifies a plurality of outlets of the intersection, and, for each outlet, identifies a polyline associated with the outlet, identifies a target point along the polyline, and determines a constant curvature path from the object to the target point. The system determines a score associated with each outlet based at least in part on the constant curvature path of the outlet, generates a pruned set of outlets that includes one or more of the outlets from the plurality of outlets based on its score, and for each outlet in the pruned set, generates a reference path from the object to the target point of the outlet.

Abstract: Systems and methods for predicting actions of an actor before entering a conflicted area are disclosed. The methods may include detecting the presence of an actor in an environment of an autonomous vehicle while the autonomous vehicle and the actor are approaching the conflicted area, determining whether the autonomous vehicle has precedence over the actor for traversing the conflicted area, assigning a kinematic target to the moving object that requires the moving object to come to a stop at an yield point before entering the conflicted area if the autonomous vehicle has precedence over the actor, determining whether a plurality of forecasted trajectories for the actor need to be generated, and controlling movement of the autonomous vehicle to traverse the conflicted area accordingly.

Abstract: A method of determining a trajectory for an autonomous vehicle is disclosed. An ego-vehicle may detect a moving actor in an environment. To choose between candidate trajectories for the ego-vehicle, the system will consider the cost of each candidate trajectory to the moving actor. The system will use the candidate trajectory costs for the candidate trajectories to select one of the candidate trajectories via which to move the ego-vehicle. An autonomous vehicle system of the ego-vehicle may then move the ego-vehicle in the environment along the selected trajectory.

Abstract: Systems and methods for operating an autonomous vehicle. The methods comprising: obtaining, by a computing device, a LiDAR dataset; plotting, by a computing device, the LiDAR dataset on a 3D graph to define a 3D point cloud; using, by a computing device, the LiDAR dataset and contents of a vector map to define a cuboid on the 3D graph that encompasses points of the 3D point cloud that are associated with an object in proximity to the vehicle, where the vector map comprises lane information; and using the cuboid to facilitate driving-related operations of the autonomous vehicle.

Abstract: Systems and methods for forecasting trajectories of objects. The method includes obtaining a prediction model trained to predict future trajectories of objects. The prediction model is trained over a first prediction horizon selected to encode inertial constraints in a predicted trajectory and over a second prediction horizon selected to encode behavioral constraints in the predicted trajectory. The method also include generating a planned trajectory of an autonomous vehicle by receiving state data corresponding to the autonomous vehicle, receiving perception data corresponding to an object, predicting a future trajectory of the object based on the perception data and the prediction model, and generating the planned trajectory of the autonomous vehicle based on the future trajectory of the object and the state data.

Abstract: A method and a system for forecasting trajectories in an autonomous vehicle using recurrent neural networks. The method includes receiving a first set of data that comprises time series information corresponding to states of a plurality of objects, analyzing the first set of data to determine a plurality of object trajectory sequences corresponding to the plurality of objects, and using one or more of the plurality of object trajectory sequences as input to train a prediction model for predicting future trajectories of the plurality of objects. The predication model can be trained by defining a first prediction horizon, training the prediction model over the first prediction horizon to generate a semi-trained prediction model, defining a second prediction horizon that is longer than the first prediction horizon, and training the semi-trained prediction model to generate a trained prediction model.

Abstract: Techniques for object re-identification based on temporal context. Embodiments extract, from a first image corresponding to a first camera device and a second image corresponding to a second camera device, a first plurality of patch descriptors and a second plurality of patch descriptors, respectively. A measure of visual similarity between the first image and the second image is computed, based on the first plurality of patch descriptors and the second plurality of patch descriptors. A temporal cost between the first image and the second image is computed, based on a first timestamp at which the first image was captured and a second timestamp at which the second image was captured. The measure of visual similarity and the temporal cost are combined into a single cost function, and embodiments determine whether the first image and the second image depict a common object, using the single cost function.

Abstract: An autonomous vehicle navigates an intersection in which occlusions block the vehicle's ability to detect moving objects. The vehicle handles this by generating a phantom obstacle behind the occlusion. The vehicle will predict the speed of the phantom obstacle and use the predicted speed to assess whether the phantom obstacle may collide with the vehicle. If a collision is a risk, the vehicle will slow or stop until it confirms that either (a) the phantom obstacle is not a real obstacle or (b) the vehicle can proceed at a speed that avoids the collision. To determine which occlusions shield real objects, the system may use a rasterized visibility grid of the area to identify occlusions that may accommodate the object.

Abstract: A method and a system for forecasting trajectories in an autonomous vehicle using recurrent neural networks. The method includes receiving a first set of data that comprises time series information corresponding to states of a plurality of objects, analyzing the first set of data to determine a plurality of object trajectory sequences corresponding to the plurality of objects, and using one or more of the plurality of object trajectory sequences as input to train a prediction model for predicting future trajectories of the plurality of objects. The predication model can be trained by defining a first prediction horizon, training the prediction model over the first prediction horizon to generate a semi-trained prediction model, defining a second prediction horizon that is longer than the first prediction horizon, and training the semi-trained prediction model to generate a trained prediction model.

Abstract: Techniques for detecting objects across images captured by camera devices. Embodiments capture, using first and second camera devices, first and second pluralities of images, respectively. First and second reference images are captured using the first and second camera devices. Color descriptors are extracted from the first plurality of images and the second plurality of images, and texture descriptors are extracted from the first plurality of images and the second plurality of images. Embodiments model a first color subspace and a second color subspace for the first camera device and the second camera device, respectively, based on the first and second pluralities of images and the first and second reference images. A data model for identifying objects appearing in images captured using the first and second camera devices is generated, based on the extracted color descriptors, texture descriptors and the first and second color subspaces.

Abstract: The disclosure provides an approach for learning latent representations of data using factorized variational autoencoders (FVAEs). The FVAE framework builds a hierarchical Bayesian matrix factorization model on top of a variational autoencoder (VAE) by learning a VAE that has a factorized representation so as to compress the embedding space and enhance generalization and interpretability. In one embodiment, an FVAE application takes as input training data comprising observations of objects, and the FVAE application learns a latent representation of such data. In order to learn the latent representation, the FVAE application is configured to use a probabilistic VAE to jointly learn a latent representation of each of the objects and a corresponding factorization across time and identity.