Robot Route Learning for Obstacle-Adaptive Motion Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot control devices struggle to ensure that autonomous traveling robots avoid obstacles in environments with varying obstacle positions and types, as machine learning devices trained in one environment may not generalize effectively to others.
Innovation Solution
A robot control device that utilizes two learning models: a first learning model to acquire a moving route based on observation data and state data, and a second learning model to generate control values for the robot to follow the acquired moving route, thereby adapting to new environments with different obstacle configurations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning device learns an operation program in a certain single environment, then the robot can move without colliding with obstacles in that specific environment, but the robot may collide with obstacles when moved to another environment with different obstacle positions or types
Solution Approach 1:
The patent divides the learning process into two distinct stages: first learning to acquire moving routes that avoid obstacles, and second learning to generate control values to execute those routes. This segmentation allows each learning model to specialize in one aspect, improving overall reliability while maintaining environmental adaptability through the modular architecture.
Solution Approach 2:
The patent transitions from single-environment learning to multi-environment learning by introducing a second learning dimension. The first learning model handles route planning across multiple environments, while the second learning model handles control execution, effectively adding a temporal and functional dimension to the learning process that enables generalization across different obstacle configurations.
2Adaptability or versatility
If a robot uses a single learning model to learn operation programs, then the device complexity is reduced, but the ability to handle varying obstacle positions and types across different environments is insufficient
Solution Approach 1:
The patent segments the single learning model into two specialized learning models: a first learning model for route acquisition and a second learning model for control value generation. This segmentation increases adaptability to different environments while managing complexity through functional specialization and modular architecture.
Solution Approach 2:
The first learning model serves as a universal route planner that can determine obstacle-free paths in various environments with different obstacle positions and types. The second learning model acts as a universal controller that executes routes regardless of environmental specifics, providing multi-functionality that enhances adaptability without proportionally increasing complexity.
Data Source
AI summary
A robot control device is configured to include: a moving route acquiring unit to acquire a moving route of a robot from a first learning model by giving, to the first learning model, observation data indicating a position of an obstacle being present in a region where the robot moves and state data indicating a moving state of the robot at a movement start point at which the robot starts moving among moving states of the robot in the region where the robot moves; and a control value generating unit to generate a control value for the robot, the control value for allowing the robot to move along the moving route acquired by the moving route acquiring unit.