公开(公告)号：US20240371127A1

公开(公告)日：2024-11-07

申请号：US18656172

申请日：2024-05-06

Applicant: Plus One Robotics, Inc.

Inventor： Nicholas Brian DePalma ,  Daniel Grollman ,  Abhijit Majumdar

IPC: G06V10/75 ,  B25J9/16 ,  G06T1/00 ,  G06T7/13 ,  G06T7/55 ,  G06T7/73 ,  G06V10/44 ,  G06V10/94 ,  G06V20/50 ,  H04N23/90

Abstract: The present disclosure is for a system and a method for computer vision based object detection. The invention uses images of objects from multiple perspectives and for each image identifies planes belonging to different objects. The planes are then analyzed to determine planes belonging to the same physical object. This is accomplished by comparing characteristics of the identified planes with each other and/or expected criteria. Planes identified as belonging to the same object can be grouped and used to provide pick instructions to a robot.

公开(公告)号：US11087172B2

公开(公告)日：2021-08-10

申请号：US17139974

申请日：2020-12-31

Applicant: Plus One Robotics, Inc.

Inventor： Jonathan Lwowski ,  Abhijit Majumdar

IPC: G06K9/62 ,  G06N3/04 ,  G06K9/00

Abstract: Training images can be synthesized in order to obtain enough data to train a model (e.g., a neural network) to recognize various classifications of a type of object. Images can be synthesized by blending images of objects labeled using those classifications into selected background images. To improve results, one or more operations are performed to determine whether the synthesized images can still be used as training data, such as by verifying one or more objects of interested represented in those images is not occluded, or at least satisfies a threshold level of acceptance. The training images can be used with real world images to train the model.

公开(公告)号：US11928594B2

公开(公告)日：2024-03-12

申请号：US17396941

申请日：2021-08-09

Applicant: Plus One Robotics, Inc.

Inventor： Jonathan Lwowski ,  Abhijit Majumdar

IPC: G06N3/08 ,  G06F18/214 ,  G06F18/2413 ,  G06N3/04 ,  G06V10/772 ,  G06V10/774 ,  G06V10/776 ,  G06V20/10

CPC classification number: G06N3/08 ,  G06F18/214 ,  G06F18/2413 ,  G06N3/04 ,  G06V10/772 ,  G06V10/774 ,  G06V10/776 ,  G06V20/10

Abstract: Training images can be synthesized in order to obtain enough data to train a model (e.g., a neural network) to recognize various classifications of a type of object. Images can be synthesized by blending images of objects labeled using those classifications into selected background images. To improve results, one or more operations are performed to determine whether the synthesized images can still be used as training data, such as by verifying one or more objects of interested represented in those images is not occluded, or at least satisfies a threshold level of acceptance. The training images can be used with real world images to train the model.

公开(公告)号：US20220203547A1

公开(公告)日：2022-06-30

申请号：US17566931

申请日：2021-12-31

Applicant: Plus One Robotics, Inc.

Inventor： Abhijit Majumdar ,  Dan Grollman ,  Zach Keeton

IPC: B25J9/16 ,  G06T1/00 ,  G06T7/50 ,  G06V10/22 ,  G06V10/46 ,  G06V20/50 ,  G06K9/62 ,  G06T7/73 ,  B25J13/08 ,  B65G59/02

Abstract: The present invention relates to pick planning for robotic picking applications to improve efficiency of automated picking operations and reduce robot down time. A pick plan is computed by obtaining data of a pick scene, processing the obtained data to identify objects and determine features associated with the objects, and determining an order and pick instructions based on the features. A computed pick plan may be periodically verified by reacquiring data of the pick scene and comparing the reacquired data with previous pick scene data in order to determine if a pick plan remains appropriate or should be updated or discarded and recomputed.

公开(公告)号：US20210201077A1

公开(公告)日：2021-07-01

申请号：US17139974

申请日：2020-12-31

Applicant: Plus One Robotics, Inc.

Inventor： Jonathan Lwowski ,  Abhijit Majumdar

IPC: G06K9/62 ,  G06K9/00 ,  G06N3/04

Abstract: Training images can be synthesized in order to obtain enough data to train a model (e.g., a neural network) to recognize various classifications of a type of object. Images can be synthesized by blending images of objects labeled using those classifications into selected background images. To improve results, one or more operations are performed to determine whether the synthesized images can still be used as training data, such as by verifying one or more objects of interested represented in those images is not occluded, or at least satisfies a threshold level of acceptance. The training images can be used with real world images to train the model.