Human-Robot Task Learning From Demonstrated Subtask Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial robots are inflexible and difficult to repurpose, requiring extensive programming knowledge and redesign when manufacturing line changes occur, limiting the ability of line workers to easily adapt robots to new tasks.
Innovation Solution
A system that records human actions, categorizes tasks into subtasks, and estimates quality to determine optimal subtask sequences for robot learning, enabling robots to adapt by observing human actions and using wearable sensors for accurate sensor readings, even in the absence of human presence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If industrial robots are programmed to perform tasks using traditional methods, then task execution reliability is improved, but adaptability to manufacturing line changes deteriorates
Solution Approach 1:
The system records human actions as demonstration data and creates a digital model of the task sequence. Instead of traditional programming, the robot learns by copying human demonstrations through sensors and cameras, storing the action sequences in a database for later execution and adaptation.
Solution Approach 2:
The patent replaces traditional mechanical programming interfaces with sensor-based perception systems. Cameras, microphones, and other sensors capture human actions and convert them into executable robot commands, substituting manual programming mechanisms with automated observation and learning systems.
2Manufacturing precision
If traditional robot programming interfaces are used, then programming precision is improved, but ease of operation deteriorates
Solution Approach 1:
The system enables line workers to programmatically train robots themselves without external integrators. The robot autonomously records, processes, and learns from human demonstrations, allowing end-users to perform programming tasks independently through simple action demonstration rather than complex coding.
Solution Approach 2:
The patent introduces an intermediate learning system that translates human actions into robot commands. This intermediary layer includes sensors, processors, and algorithms that convert natural human movements into structured robot instructions, bridging the gap between human capability and robot execution without requiring direct programming expertise.
3Manufacturing precision
If extensive programming knowledge is required for robot operation, then task execution precision is improved, but adaptability to new tasks deteriorates
Solution Approach 1:
The system breaks down complex tasks into discrete action sequences that can be independently recorded and learned. Human demonstrations are segmented into individual steps (pick, place, weld, etc.), each captured separately and stored as modular units that can be recombined and adapted for different tasks without requiring comprehensive reprogramming.
Data Source
AI summary
Example implementations described herein involve systems and methods for training and managing machine learning models in an industrial setting. Specifically, by leveraging the similarity across certain production areas, example implementations can group together these areas to train models efficiently that use human pose data to predict human activities or specific task(s) the workers are engaged in. The example implementations do away with previous methods of independent model construction for each production area and takes advantage of the commonality amongst different environments.