Method and system for grasping an object
AssigneeVicarious FPC, Inc.
InventorsAhmad Humayun, Michael Stark, Nan Rong, Bhaskara Mannar Marthi, Aravind Sivakumar
Life of this patent
Estimating the term…
Abstract
The method for increasing the accuracy of grasping an object can include: labelling an image based on an attempted object grasp by a robot and generating a trained graspability network using the labelled images. The method can additionally or alternatively include determining a grasp point using the trained graspability network; executing an object grasp at the grasp point S400; and/or any other suitable elements.
Claims21
A method comprising:
- generating training data comprising, for each of a plurality of object scenes:
- receiving an image of the object scene, wherein the object scene comprises a first plurality of objects that at least partially occludes a second plurality of objects;
- determining object parameters within the object scene using an object detector for the objects;
- determining a set of candidate grasp points based on the object parameters;
- selecting a grasp point from the set of candidate grasp points;
- and
- determining a grasp outcome for the grasp point;
- and
- training a neural network, using the image, the grasp point, and the grasp outcome for the grasp point, to output a grasp success probability map for an input image.
A method comprising:
- receiving an image of a plurality of objects;
- generating a graspability heat map comprising a grasp score for each of a set of pixels of the image;
- selecting a grasp point based on the graspability heat map, wherein the grasp point is associated with a target object;
- and
- determining control instructions for a robotic manipulator to grasp the target object based on the selected grasp point.
method of claim 15, wherein the grasp point is selected from the set of grasp points using a grasp selector which adjusts a grasp score weight based on a set of heuristics, comprising:
- a collision avoidance rule;
- and
- a failure hysteresis rule for a sliding temporal window of recent grasp attempts.
Dependencies are inferred from the claim text.