Massachusetts Institute of Technology

Nationaluniversity of Ireland, Maynooth

Near Earth Autonomy, Inc.

Michael Kaess

Sanjiv Singh

John J Leonard

Garrett Hemann

John McDonald

Jeffrey Mishler

Thomas John Whelan

On-board, computer-based systems and methods compute continuously updated, real-time state estimates for an aerial vehicle by appropriately combining, by a suitable Kalman filter, local, relative, continuous state estimates with global, absolute, noncontinuous state estimates. The local, relative, continuous state estimates can be provided by visual odometry (VO) and/or an inertial measurement unit (IMU). The global, absolute, noncontinuous state estimates can be provided by terrain-referenced navigation, such as map-matching, and GNSS. The systems and methods can provide the real-time, continuous estimates even when reliable GNSS coordinate data is not available.

Determining position and orientation for aerial vehicle in GNSS-denied situations

A method for mapping an environment comprises moving a sensor along a path from a start location (P) through the environment, the sensor generating a sequence of images, each image associated with a respective estimated sensor location and comprising a point cloud having a plurality of vertices, each vertex comprising an (x,y,z)-tuple and image information for the tuple. The sequence of estimated sensor locations is sampled to provide a pose graph (P) comprising a linked sequence of nodes, each corresponding to a respective estimated sensor location. For each node of the pose graph (P), a respective cloud slice (C) comprising at least of portion of the point cloud for the sampled sensor location is acquired. A drift between an actual sensor location (P) and an estimated sensor location (P) on the path is determined. A corrected pose graph (P') indicating a required transformation for each node of the pose graph (P) between the actual sensor location (P) and the start location (P) to compensate for the determined drift is provided. The sequence of estimated sensor locations is sampled to provide a deformation graph (N) comprising a linked sequence of nodes, each corresponding to respective estimated sensor locations along the path. For at least a plurality of the vertices in the cloud slices, a closest set of K deformation graph nodes is identified and a respective blending function based on the respective distances of the identified graph nodes to a vertex is determined. Transformation coefficients for each node of the deformation graph are determined as a function of the 20 required transformation for each node of the pose graph (P) to compensate for the determined drift. Tuple coordinates for a vertex are transformed to compensate for sensor drift as a function of the blending function and the transformation coefficients for the K deformation graph nodes closest to the vertex.

Method for mapping an environment

Growing community of inventors

Pittsburgh, PA, United States of America

Michael Kaess