Robot Force Control With Natural Force Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Robot control performance in physical human-robot interaction systems is degraded due to dynamic characteristics of users, which vary in real time, causing natural forces to be misinterpreted by force/torque sensors, leading to distorted output signals and reduced control accuracy.
Innovation Solution
An apparatus and method that includes an active force detector, a compensator, and a controller to actively compensate for natural forces by optimizing internal parameters of a predefined dynamics model, using techniques like simultaneous perturbation stochastic approximation (SPSA) and AMSGrad, to improve robot control performance by reducing the value of an energy-dependent objective function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If force/torque sensor is used to measure user applied force in pHRi system, then robot control can be implemented, but natural force from user body dynamics and robot motion distorts the measurement signal
Solution Approach 1:
The patent extracts and separates the natural force component from the total force measurement. The dynamics model specifically identifies and models the natural force generated by user body dynamics and robot motion, allowing this harmful component to be isolated and subsequently compensated for in the control signal.
Solution Approach 2:
The patent changes the parameters of the dynamics model in real-time to adapt to varying user characteristics. By continuously updating the model parameters based on observed user behavior and robot state, the system maintains accurate natural force estimation despite changes in user posture, mass, or interaction style.
2Reliability
If user dynamic characteristics are compensated for in real-time, then robot control performance is improved, but system complexity increases due to need for continuous parameter optimization
Solution Approach 1:
The patent performs preliminary action by pre-defining the dynamics model structure and parameters before actual interaction begins. The model framework, including mass, damping, and stiffness parameters, is established in advance, reducing the computational burden during real-time operation to only updating specific parameter values based on observed dynamics.
Solution Approach 2:
The patent implements continuous feedback by using the measured force/torque signal and robot state to continuously update the dynamics model parameters. This closed-loop approach allows the system to adapt to changing user characteristics in real-time, maintaining high control performance without requiring complex reconfiguration.
3Measurement precision
If dynamics model parameters are optimized to reduce objective function value, then compensation accuracy is improved, but computational time increases
Solution Approach 1:
The patent applies partial action by optimizing only the critical dynamics model parameters (mass, damping, stiffness) that have the greatest impact on natural force estimation, rather than attempting to optimize all possible parameters. This selective approach achieves sufficient compensation accuracy while minimizing computational time.
Solution Approach 2:
The patent efficiently manages parameter optimization by changing only the necessary parameters in real-time based on observed user characteristics and robot state. Less critical parameters are pre-set and updated less frequently, reducing the overall computational burden while maintaining adequate compensation accuracy.
Data Source
AI summary
Provided are an apparatus and method for controlling a robot. The apparatus includes an active force detector configured to detect an active force, to which a natural force caused by a physical interaction between a user and a robot and not reflecting an operation intention of the user is applied, applied by the user to the robot operating through the physical interaction with the user, a compensator configured to determine a compensation force for actively compensating for the natural force applied to the active force by using a method of optimizing an internal parameter of a predefined dynamics model, and a controller configured to determine an operation instruction for controlling an operation of the robot from a result obtained by applying the compensation force determined by the compensator to the active force detected by the active force detector and operate the robot.