Robot Force Sensor Calibration for Gravity Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing calibration techniques for contact force in robots using force sensors face errors due to gravity components, especially when the robot arm's posture changes, as they rely on mass point models that fail to accurately estimate these components.
Innovation Solution
A calibration device that includes a position information acquiring unit, a force information acquiring unit, a first estimating unit using a physical model, and a second estimating unit employing a neural network to learn and correct the differences between estimated and detected forces, effectively suppressing estimation errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a mass point model is used to estimate gravity components, then the calibration process is simplified, but estimation accuracy deteriorates due to errors in modeling mass distribution and posture-dependent gravity changes
Solution Approach 1:
The patent replaces the traditional mass point model (mechanical/mathematical modeling approach) with a neural network model (data-driven approach). The neural network learns the relationship between robot posture and gravity components from training data, automatically capturing complex mass distribution effects and posture-dependent variations without requiring explicit mathematical modeling. This substitution resolves the contradiction by achieving high estimation accuracy through learning while maintaining relatively simple implementation.
Solution Approach 2:
The patent transforms the calibration problem from estimating physical parameters (mass distribution, centroid position) to learning a mapping function from posture to gravity components. By changing from parameter-based estimation to function-based prediction using neural networks, the system achieves higher accuracy for posture-dependent gravity variations without increasing device complexity.
2Device complexity
If force sensor detection results are used directly to determine contact force, then the measurement process is simplified, but measurement accuracy deteriorates due to inclusion of gravity components
Solution Approach 1:
The patent extracts and removes the gravity component from the force sensor detection results. By using the neural network to estimate gravity components separately and subtracting them from the total force measurement, the system isolates the contact force. This extraction approach resolves the contradiction by maintaining simple force measurement while achieving accurate contact force determination.
Solution Approach 2:
The patent introduces the neural network as an intermediary component that processes the relationship between posture and gravity components. This intermediary enables the separation of gravity effects from contact forces without requiring direct modification of the force sensor or complex mechanical adjustments, thus maintaining measurement simplicity while improving accuracy.
Data Source
AI summary
Disclosed is a calibration device including: a position information acquiring unit (101) for acquiring position information showing the position and the posture of control target equipment; a force information acquiring unit (102) for acquiring information about a force applied to the control target equipment from a detection result of a force sensor (5) disposed in the control target equipment; a first estimating unit (104) for estimating the force applied to the control target equipment from the acquired position information by using a physical model, to acquire estimated force information; and a second estimating unit (105) for estimating a linear or nonlinear model on the basis of the acquired position information, the acquired force information, and the acquired estimated force information.