Back to results
US 11,446,816 B2Granted
Robotic systems and methods for robustly grasping and targeting objects
AssigneeThe Regents of the University of California
InventorsKenneth Yigael Goldberg, Jeffrey Brian Mahler, Matthew Matl
Life of this patent
Estimating the term…
Abstract
Embodiments are generally directed to generating a training dataset of labelled examples of sensor images and grasp configurations using a set of three-dimensional (3D) models of objects, one or more analytic mechanical representations of either or both of grasp forces and grasp torques, and statistical sampling to model uncertainty in either or both sensing and control. Embodiments can also include using the training dataset to train a function approximator that takes as input a sensor image and returns data that is used to select grasp configurations for a robot grasping or targeting mechanism.
Claims23
A computer-implemented method, comprising:
- generating a training dataset of labelled examples of sensor images and grasp configurations using:
- a set of three-dimensional (3D) models of objects;
- one or more analytic mechanical representations of either or both of grasp forces and grasp torques;
- and
- statistical sampling to model uncertainty in either or both sensing and control, the generated training dataset including a measure of grasp quality associated with a grasp configuration, wherein the statistical sampling includes uncertainty in variables related to initial state and contact area;
- using the training dataset to train a Convolutional Neural Network (CNN) that takes as input a sensor image and generates a set of possible grasp configurations that are available for a robot grasping mechanism that includes one or more suction cups;
- evaluating a level of robustness for each of the set of possible grasp configurations;
- and
- selecting one of the set of possible grasp configurations that has a highest level of robustness.
6 more elements
A computer-implemented method, comprising:
- generating a training dataset of labelled examples of sensor images and target points within those images using:
- a set of three-dimensional (3D) models of objects;
- one or more analytic evaluation methods of desired target points on an object;
- and
- statistical sampling to model uncertainty in either or both sensing and control, wherein the statistical sampling includes uncertainty in variables related to at least one selected from a group consisting of:
- initial state, contact, physical motion, friction, inertia, object shape, robot control, and sensor data;
- and
- using the training dataset to train a Convolutional Neural Network (CNN) that takes as input a sensor image and returns data to compute a single target point for a robot targeting mechanism that includes at least one selected from a group consisting of:
- placing a label on the object, affixing a stamp to the object, and inspecting the object.
6 more elements
An apparatus, comprising:
- a sensor;
- a robotic grasping mechanism that includes one or more suction cups;
- and
- one or more processors configured to use sensor images to compute a desired grasp configuration for the robotic grasping mechanism based at least in part on a Convolutional Neural Network (CNN) that is trained on a training dataset of labelled examples of sensor images and grasp configurations using:
- a set of three-dimensional (3D) models of objects;
- one or more analytic mechanical representations of either or both of grasp forces and grasp torques;
- and
- statistical sampling to model uncertainty in either or both sensing and control;
- wherein the CNN is configured to take as input a sensor image and generate a set of possible grasp configurations that are available for the robotic grasping mechanism, evaluate a level of robustness for each of the set of possible grasp configurations, and select one of the set of possible grasp configurations that has a highest level of robustness.
6 more elements
An apparatus, comprising:
- a sensor;
- a robotic targeting mechanism;
- and
- one or more processors configured to use sensor images to compute a desired target configuration for the robotic targeting mechanism based at least in part on a Convolutional Neural Network (CNN) that is trained on a training dataset of labelled examples of sensor images and target points using:
- a set of three-dimensional (3D) models of objects;
- one or more analytic mechanical representations of target points;
- and
- statistical sampling to model uncertainty in either or both of sensing and control, wherein the statistical sampling includes uncertainty in variables related to initial state and contact area;
- wherein the function approximator is configured to take as input a sensor image and generate a single robust target point for the robotic targeting mechanism by evaluating a level of robustness for each of a set of possible grasp configurations and selecting one of the set of possible grasp configurations that has a highest level of robustness.
6 more elements
Dependencies are inferred from the claim text.