Back to results
US 12,390,929 B2Granted
Object rearrangement using learned implicit collision functions
AssigneeNVIDIA Corporation
InventorsMichael Danielczuk, Arsalan Mousavian, Clemens Eppner, Dieter Fox
Life of this patent
Estimating the term…
Abstract
Apparatuses, systems, and techniques for determining whether collisions will occur in potential paths of an object within a scene. In at least one embodiment, one or more neural networks determine whether collisions will occur in potential paths of an object within a scene based at least in part on point cloud data of the object and the scene.
Claims29
A method to detect whether a collision with an object in a scene will occur, the method comprising:
- determining a set of object features based at least in part on a first point cloud associated with the object;
- determining a set of scene features based at least in part on a second point cloud associated with the scene;
- predicting a collision-free object path of the object within the scene at least by:
- generating one or more queries based at least in part on the set of object features, the set of scene features, and one or more transforms associated with the object, the one or more queries to comprise one or more potential object paths through the scene and one or more object rotations along at least one of the one or more potential object paths;
- and
- using at least one neural network to infer the collision-free object path, based at least in part on input of the one or more queries into the at least one neural network, wherein the at least one neural network comprises one or more classifier neural networks;
- and
- causing at least one device to move the object along the collision-free object path from a first position to a second position.
5 more elements
A non-transitory computer-readable storage medium having stored thereon executable instructions that, as a result of being executed by one or more processors of a computer system, cause the computer system to at least:
- obtain a set of point clouds indicating at least a scene and an object;
- determine a set of object features based at least in part on a first point cloud in the set of point clouds, the first point cloud to be associated with the object;
- determine a set of scene features based at least in part on a second point cloud in the set of point clouds, the second point cloud to be associated with the scene;
- determine a collision-free object path of the object within the scene at least by:
- generating one or more queries based at least in part on the set of object features, the set of scene features, and one or more transforms associated with the object, the one or more queries to comprise one or more potential object paths through the scene and one or more object rotations along at least one of the one or more potential object paths;
- and
- predicting, using at least one neural network, the collision-free object path, based at least in part on input of the one or more queries into the at least one neural network, wherein the at least one neural network comprises one or more classifier neural networks;
- and
- cause at least one device to move the object along the predicted collision-free object path from a first position to a second position.
6 more elements
A system, comprising:
- one or more computers having one or more processors to use one or more neural networks, wherein the one or more neural networks comprise at least one classifier neural network, the one or more processors of the one or more computers to:
- determine a set of object features based at least in part on a first point cloud associated with an object,
- determine a set of scene features based at least in part on a second point cloud associated with a scene,
- determine a path of the object within the scene at least by:
- generate one or more queries based at least in part on the set of object features, the set of scene features, and one or more transforms, the one or more queries to comprise one or more potential object paths comprising one or more object rotations along at least one of the one or more potential object paths, and
- select the path from the one or more potential object paths, based at least in part on input of the one or more queries into the one or more neural networks, including the at least one classifier neural network, to classify the one or more potential object paths, to indicate one or more likelihoods that the one or more potential object paths will result in the object colliding with the scene;
- and
- cause at least a portion of a device to move from a first position to a second position in accordance with the selected path.
5 more elements
A system comprising one or more processors to:
- determine a set of object features based at least in part on a first point cloud associated with an object;
- determine a set of scene features based at least in part on a second point cloud associated with a scene;
- determine a path of the object within the scene at least by:
- generating one or more queries based at least in part on the set of object features, the set of scene features, and one or more transforms associated with the object, the one or more queries to comprise one or more potential object paths through the scene and one or more object rotations along at least one of the one or more potential object paths;
- and
- predicting, using at least one neural network, the path of the one or more potential object paths, based at least in part on input of the one or more queries into the at least one neural network, wherein the at least one neural network comprises one or more classifier neural networks to be used to classify the one or more potential object paths with one or more probabilities of colliding with the scene;
- and
- cause at least one device to move the object along the predicted path from a first position to a second position.
5 more elements
Dependencies are inferred from the claim text.