Robot-Carried Object Pose Referencing After Grasp Shift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotic systems struggle with accurately determining the pose of objects after they have been grasped, leading to inaccuracies and potential collisions due to unpredictable shifting during and after grasping, which complicates tasks requiring precise object positioning and can compromise safety.
Innovation Solution
The system employs a method that includes using information about an expected object pose, repositioning the object for perception data gathering, and processing estimates from a neural network to generate accurate object references, utilizing three-dimensional information to register a known model with real-time pose data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the robot grasps the object, then the object can be moved to the desired location, but the object may shift unpredictably during and after grasping, leading to pose estimation inaccuracy
Solution Approach 1:
The system performs preliminary pose estimation before grasping the object, establishing an expected pose reference. This preliminary action allows the system to anticipate the object's initial position and orientation, which is then used to guide subsequent perception and correction processes after grasping occurs.
Solution Approach 2:
The system continuously gathers perception data about the object's actual pose after grasping and compares it with the expected pose. This feedback loop enables real-time detection of pose deviations caused by object shifting, allowing the system to generate corrected pose estimates and adjust its actions accordingly to maintain accuracy.
2Measurement precision
If the robot repositions the object for perception data gathering, then more accurate object references can be generated, but additional time is required for the repositioning process
Solution Approach 1:
The system performs only the necessary repositioning actions required to obtain sufficient perception data, rather than completely repositioning the object. This partial action approach gathers adequate information for accurate pose estimation while minimizing the time and effort required, avoiding excessive manipulation of the object.
3Measurement precision
If the robot uses sophisticated perception and processing methods, then object pose accuracy improves, but the system complexity increases
Solution Approach 1:
The perception and processing system is divided into distinct functional modules: perception data gathering, neural network processing, pose estimation, and correction. Each module performs a specific function, making the overall complex system more manageable and maintainable while achieving high accuracy through coordinated operation of specialized components.
Solution Approach 2:
The system introduces an intermediary expected pose reference that mediates between the initial object state and the final corrected pose estimation. This intermediary serves as a reference framework that simplifies the processing of perception data by providing a baseline for comparison and correction, reducing the computational complexity of direct pose estimation.
Data Source
AI summary
A method in accordance with a particular embodiment of the present technology includes retrieving an object from a starting location by grasping the object between two end effectors of a mobile robot. The method further includes generating, by data-processing hardware operably associated with the mobile robot, an object reference corresponding to a pose of the object at a time after retrieving the object and while the object is in contact with the end effectors. Generating the object reference is based at least partially on perception data corresponding to the object and includes generating a three-dimensional reference corresponding to a feature of the object based at least partially on a two-dimensional estimate from a neural network and a three-dimensional reference corresponding to a different feature of the object. Finally, the method includes controlling the mobile robot based at least partially on the object reference.