Machine Learning Pose Correction for Robotic Grasping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotics planning requires immense manual programming for precise robotic component movements, especially for grasping objects in the correct orientation and position, which is tedious, error-prone, and often necessitates specialized fixture holders or low-quality image-based estimation, adding burden and expense.
Innovation Solution
A system trains a machine learning model to map images of an object to data representing the offset between the pose of the robotic component and a nominal pose, allowing for flexible and efficient pose correction without specialized fixtures, using optical and tactile images to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional manual programming is used to precisely control robotic component movements, then manufacturing precision is improved, but device complexity and time consumption increase significantly
Solution Approach 1:
The patent replaces manual programming and mechanical positioning systems with a machine learning model that processes images to directly predict pose offsets. The system substitutes complex mechanical control workflows with an intelligent vision-based system that automatically corrects robotic component poses by mapping image inputs to pose correction outputs through trained neural networks.
Solution Approach 2:
The robotic system performs self-correction by using its own captured images to identify and correct pose deviations. The machine learning model enables the system to autonomously determine corrections without external intervention or complex pre-programming, allowing the robotic component to self-adjust its pose based on visual feedback from the environment.
2Manufacturing precision
If specialized fixture holders are used to ensure correct grasping orientation and position, then manufacturing precision is improved, but device complexity and adaptability worsen due to requiring custom fixtures for each part
Solution Approach 1:
The patent implements a universal image-based pose correction system that can handle multiple different parts and robotic components without requiring specialized fixtures for each. The machine learning model is trained on diverse datasets and can generalize to new part types, providing a single versatile solution that replaces the need for part-specific mechanical fixtures while maintaining high grasping precision.
Solution Approach 2:
The system replaces physical specialized fixtures with a virtual vision-based correction system. Instead of using mechanical fixture holders to physically constrain and position parts, the system uses image processing and machine learning to calculate and apply pose corrections, eliminating the need for physical adaptation hardware for different part types.
3Device complexity
If image-based pose estimation is used after grasping, then device complexity is reduced, but measurement precision worsens due to occlusion by the robotic component
Solution Approach 1:
The patent performs pose correction before the robotic component executes its primary task, using images captured during the approach phase when occlusion is minimal. The machine learning model processes these preliminary images to predict pose offsets and generate correction commands that are applied before final grasping, ensuring high measurement precision while maintaining system simplicity.
Solution Approach 2:
Instead of estimating pose after grasping when occlusion occurs, the system inverts the approach by using images taken during the approach to pre-calculate pose corrections. The machine learning model processes approach-phase images to determine corrections that will be applied before final positioning, thereby avoiding the occlusion problem entirely while maintaining measurement precision.
4Measurement precision
If the robotic component moves to camera range and stops for imaging, then measurement precision is improved, but productivity decreases due to added time and movement complexity
Solution Approach 1:
The patent maintains continuous robotic motion by capturing images during the approach phase without requiring the robotic component to stop or reposition to camera range. The machine learning model processes these continuous-action images to generate pose corrections that are applied on-the-fly, eliminating idle time and maintaining productivity while achieving sufficient measurement precision through the intelligent processing capability.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing pose correction. One of the methods includes receiving a plurality of images of an object held by a robotic component, wherein each image is associated with a respective perturbation pose of the robotic component that is at an offset relative to a nominal pose; generating a plurality of training examples, wherein each training example includes one or more of the plurality of images and data representing an offset between a perturbation pose associated with the one or more images and the nominal pose; and training a machine learning model that is configured to map an input comprising one or more images of an object to an output comprising data representing the offset between the perturbation pose of the robotic component and the nominal pose.