Machine Learning Object Orientation With Impulse Actuator Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotic systems lack flexibility in orienting objects of varying shapes and sizes, requiring reconfiguration of hard automation for each part, which is time-consuming and labor-intensive, and are unable to efficiently orient objects in any desired pose, especially in mass customization scenarios.
Innovation Solution
A machine learning-based control system using an array of actuators to apply impulses on objects, with a closed-loop controller and imaging system to determine the optimal actuator commands for changing the object's orientation to a desired pose without requiring detailed modeling of contact and impulse dynamics, allowing for self-supervised learning and adaptation to different objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If hard automation and single-part orientation systems are used, then manufacturing precision for specific parts is improved, but adaptability to different objects is worsened
Solution Approach 1:
The system uses a camera to capture images of objects regardless of their shape, size, or orientation, and the machine learning model processes these images to determine orientation commands for any object type, making the system universal rather than dedicated to specific parts
Solution Approach 2:
The patent replaces traditional mechanical orientation devices (vibratory bowl feeders, tilt tables, flip tables) with a vision-based detection system and machine learning-based control system that uses actuators to orient objects, eliminating the need for mechanical reconfiguration
2Manufacturing precision
If custom programs are created for each desired orientation, then orientation accuracy is improved, but device complexity and time consumption are worsened
Solution Approach 1:
The machine learning model automatically learns the mapping between object images and orientation commands from training data, enabling the system to self-adapt to different objects without requiring manual programming or reconfiguration for each object type
Solution Approach 2:
The system changes the approach from fixed mechanical configurations to variable software-based control, where the machine learning model adjusts its predictions based on the input image characteristics, enabling flexible adaptation without physical reconfiguration
3Measurement precision
If traditional physical methods are used to model contact and impulse dynamics, then theoretical accuracy is improved, but practical feasibility is worsened
Solution Approach 1:
The patent replaces complex physical dynamics modeling with a machine learning approach that directly maps object images to orientation commands, avoiding the need to model contact and impulse dynamics while achieving practical orientation results
Solution Approach 2:
Instead of modeling the complex physical dynamics of object manipulation, the system uses image copying and pattern recognition through machine learning to predict the required orientation commands, simplifying the problem from physics-based modeling to visual pattern matching
4Adaptability or versatility
If reconfiguration of hard automation is performed for each object type, then adaptability is improved, but productivity is worsened
Solution Approach 1:
The system transitions from static mechanical configurations to dynamic software-based control, where the machine learning model can rapidly adapt to different objects through image processing without requiring physical reconfiguration, enabling fast changeover between object types
Solution Approach 2:
The machine learning model is pre-trained with a diverse dataset of object images and corresponding orientation commands, enabling it to quickly adapt to new object types during production without requiring time-consuming reconfiguration or reprogramming
Data Source
AI summary
A robotic controller controls orientation of an object in a desired orientation. The controller obtains pose data indicative of a location and an orientation of the object on a supporting surface and determines one or more control commands for actuating actuators, corresponding to the location and orientation of the object on the supporting surface. The actuators are activated according to the one or more control commands to apply impulse forces to the supporting surface with a likelihood of changing the orientation of the object to the desired orientation. The controller iteratively repeats these procedures until the object is oriented in the desired orientation.