Robot Motion Learning with MPC-Based Constraint Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional model parameter learning methods for mobile robots face issues with convergence and controllability due to differences in movement constraints between the robot and the behavior trainer, leading to ineffective movement in crowded environments and potential failure in learning the movement trajectory.
Innovation Solution
A model parameter learning method that synchronizes and synchronizes the time series of surrounding environment information and movement trajectory of a second moving body with the first moving body, using a model predictive control algorithm to calculate discrete learning speed commands that reflect the movement constraints of the first moving body, and employs a machine learning algorithm to learn the model parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a pedestrian is used as a behavior trainer for learning model parameters, then the learning model can be trained with real movement data, but the mobile robot cannot appropriately move in a crowd due to different movement constraints between the robot and pedestrian
Solution Approach 1:
The patent introduces an intermediary processing step that transforms pedestrian movement data into robot-appropriate movement data. A movement constraint conversion unit converts the pedestrian's movement trajectory into a form that respects the robot's movement constraints, acting as a mediator between the behavior trainer's data and the robot's control system.
Solution Approach 2:
The patent changes the parameters of the movement data to match the robot's constraints. The movement constraint conversion unit modifies parameters such as speed, acceleration, and trajectory to ensure the learned behavior complies with the robot's physical limitations while preserving the essential movement patterns.
2Measurement precision
If continuous value such as speeds of a pedestrian is used as learning data, then the movement trajectory can be accurately represented, but convergence of learning may be deteriorated and learning of model parameters may not converge
Solution Approach 1:
The patent transforms continuous speed values into discrete speed command values. The movement constraint conversion unit quantizes the continuous speed data into discrete levels that are suitable for the robot's control system, improving learning convergence while maintaining sufficient movement accuracy.
Solution Approach 2:
The patent uses discrete speed command values as simplified representations of continuous movement data. These discrete values act as simplified proxies that capture the essential movement information without the computational complexity of continuous values, enabling more reliable learning convergence.
Data Source
AI summary
Provided is a model parameter learning method by which a model parameter of a learning model used in control of a moving body having movement constraints can be appropriately learned. In this model parameter learning method, a model prediction control algorithm reflecting movement constraints of a robot 1 is used to calculate a time series of learning speed commands such that the movement trajectory of the robot 1 tracks the time series of a movement trajectory of a first pedestrian 5; and a model parameter of a CNN model is learned by an error back propagation method, the CNN model using learning data including the learning speed commands time series as input and outputting a time series of speed commands for a first moving body.