Ford Global Technolgoies, LLC

Gaurav Pandey

Punarjay Chakravarty

Praveen Narayanan

Mostafa Parchami

Shubham Shrivastava

Nikita Jaipuria

Sarah Houts

Dimitar Petrov Filev

Jinesh J Jain

Enrique Corona

James McBride

Graham Mills

Tyler Reid

Tom Roussel

Jianbo Lu

Shane Elwart

Yifan Chen

Gintaras Vincent Puskorius

Jeffrey Thomas Remillard

Bruno Sielly Jales Costa

John Anthony Lockwood

Justin Miller, Jr

Helen Elizabeth Kourous-Harrigan

Juan Enrique Castorena Martinez

Andrew P Richardson

Gautham Sholingar

Shreyasha Paudel

Lynn Valerie Keiser

Jinhyoung Oh

Brian George Buss

Kunjan Singh

Donna L Bell

Ganesh Kumar

Weifeng Xiong

Sagar Manglani

Devarth Parikh

Sushruth Nagesh

Bhushan Ghadge

Matthew Baer Wilmes

Xue Iuan Wong

Subodh Mishra

Alexandra Carlson

Ramanarayan Vasudevan

Rahul Kumar Agrawal

Spencer Carmichael

Katherine Skinner

A computer that includes a processor and a memory, the memory including instructions executable by the processor to generate a current keyframe point cloud based on pairs of stereo images while a stereo camera travels through a scene to determine a similar viewpoint query matrix and an opposing viewpoint query matrix based the current keyframe point cloud. A distance matrix and an opposing view distance matrix can be generated by comparing the similar viewpoint query matrix and the opposing viewpoint query matrix to reference matrices. A relative pose between a stereo camera and a reference can be determined to determine a location in the scene during travel of the stereo camera through the scene by performing sequence matching in the distance matrix and the opposing view distance matrix to determine a minimum sequence.

System for operating a movable device

A robot can be moved in a structure that includes a plurality of downward-facing cameras, and, as the robot moves, upward images can be captured with an upward-facing camera mounted to the robot. Downward images can be captured with the respective downward-facing cameras. Upward-facing camera poses can be determined at respective times based on the upward images. Further, respective poses of the downward-facing cameras can be determined based on (a) describing motion of the robot from the downward images, and (b) the upward-facing camera poses determined from the upward images.

Infrastructure to support vehicle navigation

A first plurality of center points of first two-dimensional bounding boxes corresponding to a vehicle occurring in a first plurality of images acquired by a first camera can be determined. A second plurality of center points of second two-dimensional bounding boxes corresponding to the vehicle occurring in a second plurality of images acquired by a second camera can also be determined. A plurality of non-linear equations based on the locations of the first and second pluralities of center points and first and second camera parameters corresponding to the first and second cameras can be determined. The plurality of non-linear equations can be solved simultaneously for the locations of the vehicle with respect to the first and second cameras and the six degree of freedom pose of the second camera with respect to the first camera.

Camera calibration

A trailer angle identification system includes an imaging device configured to capture an image. A controller is configured to receive steering angle data corresponding to a steering angle of a vehicle, receive vehicle speed data corresponding to a vehicle speed, and estimate an angle of the trailer relative to the vehicle by processing the image, the steering angle data, and the vehicle speed data in at least one neural network.

Systems and methods for trailer angle detection

An image can be input to a deep neural network to determine a point in the image based on a center of a Gaussian heatmap corresponding to an object included in the image. The deep neural network can determine an object descriptor corresponding to the object and include the object descriptor in an object vector attached to the point. The deep neural network can determine object parameters including a three-dimensional location of the object in global coordinates and predicted pixel offsets of the object. The object parameters can be included in the object vector, and the deep neural network can predict a future location of the object in global coordinates based on the point and the object vector.

Three-dimensional object detection

A camera is positioned to obtain an image of an object. The image is input to a neural network that outputs a three-dimensional (3D) bounding box for the object relative to a pixel coordinate system and object parameters. Then a center of a bottom face of the 3D bounding box is determined in pixel coordinates. The bottom face of the 3D bounding box is located in a ground plane in the image. Based on calibration parameters for the camera that transform pixel coordinates into real-world coordinates, a) a distance from the center of the bottom face of the 3D bounding box to the camera relative to a real-world coordinate system and b) an angle between a line extending from the camera to the center of the bottom face of the 3D bounding box and an optical axis of the camera are determined. The calibration parameters include a camera height relative to the ground plane, a camera focal distance, and a camera tilt relative to the ground plane. A six degree-of-freedom (6DoF) pose for the object is determined based on the object parameters, the distance, and the angle.

Neural network object pose determination

First feature points can be determined which correspond to pose-invariant surface model properties based on first data points included in a first lidar point cloud acquired by a sensor. A three-dimensional occupancy grid can be determined based on first data points included in the first lidar point cloud. Dynamic objects in a second lidar point cloud acquired by the sensor can be determined based on the occupancy grid. Second feature points can be determined which correspond to pose-invariant surface model properties based on second data points included in the second lidar point cloud not including the dynamic objects. A difference can be determined between corresponding feature points included in the first feature points and the second feature points. A traffic infrastructure system can be alerted based on the difference exceeding a threshold.

Sensor localization

A depth image of an object can be input to a deep neural network to determine a first four degree-of-freedom pose of the object. The first four degree-of-freedom pose and a three-dimensional model of the object can be input to a silhouette rendering program to determine a first two-dimensional silhouette of the object. A second two-dimensional silhouette of the object can be determined based on thresholding the depth image. A loss function can be determined based on comparing the first two-dimensional silhouette of the object to the second two-dimensional silhouette of the object. Deep neural network parameters can be optimized based on the loss function and the deep neural network can be output.

Object pose estimation

An image including a first object can be input to a deep neural network trained to detect objects. The deep neural network can output a first feature vector corresponding to the first object. A first distance can be measured from the first feature vector to a feature vector subspace determined using a k-means single value decomposition algorithm on an overcomplete dictionary of feature vectors. The first object can be determined to correspond to an anomaly based on the first distance.

Anomaly detection for deep neural networks

A computer includes a processor and a memory storing instructions executable by the processor to receive a time series of vectors from a sensor, determine a weighted moving mean of the vectors, determine an inverse covariance matrix of the vectors, receive a current vector from the sensor, determine a squared Mahalanobis distance between the current vector and the weighted moving mean, and output an indicator of an anomaly with the sensor in response to the squared Mahalanobis distance exceeding a threshold. The squared Mahalanobis distance is determined by using the inverse covariance matrix.

Anomaly detection in multidimensional sensor data

A method for controlling a robotic vehicle in a delivery environment includes causing the robotic vehicle to deploy from an autonomous vehicle (AV) at a first AV position in the delivery environment. The method further includes localizing, via a robotic vehicle controller, an initial position within a global reference map using a robot vehicle perception system, receiving, from the AV, a 3-dimensional (3D) augmented map and localizing an updated position in the delivery environment based on the 3D augmented map and the global reference map. The robot vehicle perception system senses obstacle characteristics, and generates a unified 3D augmented map with robot-sensed obstacle characteristics. The method further includes generating a dynamic path plan to a package delivery destination using the unified 3D augmented map, and actuating the robot vehicle to the package delivery destination according to the dynamic path plan.

Autonomous vehicle and infrastructure aided robotic system for end-to-end package delivery

A first six degree-of-freedom (DoF) pose of an object from a perspective of a first image sensor is determined with a neural network. A second six DoF pose of the object from a perspective of a second image sensor is determined with the neural network. A pose offset between the first and second six DoF poses is determined. A first projection offset is determined for a first two-dimensional (2D) bounding box generated from the first six DoF pose. A second projection offset is determined for a second 2D bounding box generated from the second six DoF pose. A total offset is determined by combining the pose offset, the first projection offset, and the second projection offset. Parameters of a loss function are updated based on the total offset. The updated parameters are provided to the neural network to obtain an updated total offset.

Neural network object detection

An extrinsic-calibration system includes at least one target and a computer. Each target includes three flat surfaces that are mutually nonparallel and a corner at which the three surfaces intersect. The computer is programmed to estimate a first set of relative positions of the corner in a first coordinate frame from a first sensor, the first set of relative positions corresponding one-to-one to a set of absolute positions of the at least one target; estimate a second set of relative positions of the corner in a second coordinate frame from a second sensor, the second set of relative positions corresponding one-to-one to the set of absolute positions; and estimate a rigid transformation between the first coordinate frame and the second coordinate frame based on the first set of relative positions and the second set of relative positions.

Sensor calibration

A set of first candidate topologies of first candidate roadside infrastructure nodes at respective mounting locations in a geographic area is randomly generated. For each of the first candidate topologies, first simulations, including detection of objects according to selected sensor parameters, installation parameters, and environment parameters for the candidate nodes at the respective mounting locations, are executed. First fitness scores are determined for each of the first candidate topologies by comparing results of the first simulations to ground truth data. Upon identifying one of the first fitness scores as exceeding a threshold, the candidate topology associated with the identified first fitness score is identified for deployment.

Roadside infrastructure deployment

Example localization systems and methods are described. In one implementation, a method receives a camera image from a vehicle camera and cleans the camera image using a VAE-GAN (variational autoencoder combined with a generative adversarial network) algorithm. The method further receives a vector map related to an area proximate the vehicle and generates a synthetic image based on the vector map. The method then localizes the vehicle based on the cleaned camera image and the synthetic image.

Localization systems and methods

Growing community of inventors

Foster City, CA, United States of America

Gaurav Pandey