Robotic Assembly Policies for High-Mixture Part Variations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotic assembly systems often require customized and predefined fixtures, tooling, and waypoints, which limits their applicability in high-mixture settings where a robot is required to assemble many different types of parts, each potentially varying in shape, size, and orientation, and the disclosed techniques enable robotic assembly in high-mixture settings where a robot is required to assemble many different types of parts, each potentially varying in shape, size, and orientation, and the ability to adapt movements based on real-time part observations rather than adhering to rigid pre-programmed waypoints.
Innovation Solution
The disclosed techniques involve generating, using, and reversing expert assembly data to train a machine learning model to control a robot to perform specific robotic assembly tasks, which include generating, and reversing assembly trajectories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If predefined sequences and manually engineered pipelines are used to guide the joining of parts, then the assembly process can be divided into distinct modules for handling parts, aligning parts, and inserting parts, but the system requires customized and predefined fixtures, tooling, and waypoints which limits applicability in high-mixture settings
Solution Approach 1:
The patent replaces manually engineered pipelines and predefined sequences with machine learning models that learn assembly trajectories from demonstrated data. The system substitutes rigid mechanical fixtures and waypoints with learned policies that can adapt to different part geometries and assembly requirements, enabling the robot to generalize across high-mixture settings without customized tooling
Solution Approach 2:
The patent changes the parameters of the assembly system from fixed predefined values to learned continuous parameters. The machine learning models learn optimal trajectories, forces, and motion parameters from demonstrated assembly data, allowing the system to adapt to varying part shapes, sizes, and orientations in high-mixture environments without requiring reconfiguration of fixtures or tooling
2Reliability
If a robot is configured to place one specific shape of a part at a fixed waypoint, then the assembly task can be completed with predefined trajectories, but the robot struggles or requires extensive reconfiguration when asked to place differently sized or shaped parts
Solution Approach 1:
The patent creates a universal robot controller that can handle multiple part types and assembly tasks using the same machine learning model. The system learns from diverse demonstrated data to become multi-functional, capable of placing different shaped parts at appropriate locations without reconfiguration, thereby achieving both reliability and versatility
Solution Approach 2:
The patent uses demonstration data as templates for learning assembly skills. Instead of programming specific trajectories for each part type, the system copies successful assembly behaviors from demonstrated data and generalizes them to new part variations, enabling reliable placement across different geometries through learned patterns rather than predefined waypoints
Data Source
AI summary
The disclosed method for training a machine learning model to control a robot includes performing, based on demonstration data associated with one or more assembly tasks, one or more first training operations to generate one or more first trained machine learning models, wherein each first trained machine learning included in the one or more first trained machine learning models is trained to control a robot to perform a different assembly task, and performing, based on the one or more first trained machine learning models and one or more geometries associated with one or more parts, one or more second training operations to generate a second trained machine learning model, wherein the second trained machine learning model is trained to control the robot to perform a plurality of assembly tasks.