Robot Torque Modeling With Friction and Residual Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot control methods rely heavily on feedback control, which can lead to stiff movements and increased risk of injury, especially when encountering obstacles, due to inaccuracies in rigid body models that fail to account for friction and other dynamic effects.
Innovation Solution
A hybrid model combining a rigid body model, a friction model, and a machine learning model is used to accurately map parameters of connection points onto torques for actuators, allowing for precise and compliant movement by accounting for friction and residual torque contributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a rigid body model is used for robot control, then the control structure is simple, but the model accuracy is insufficient leading to stiff movements and increased injury risk
Solution Approach 1:
The patent combines three distinct models (rigid body model, friction model, and machine learning model) into a unified hybrid model. The rigid body model provides the fundamental dynamic framework, the friction model accounts for dissipative effects, and the machine learning model captures residual nonlinearities. This merging resolves the contradiction by integrating multiple modeling approaches to achieve both structural organization and high accuracy, eliminating stiff movements while maintaining control effectiveness.
Solution Approach 2:
The hybrid model functions as a composite modeling approach, where each component model contributes specific strengths. The rigid body model handles inertial effects, the friction model addresses velocity-dependent losses, and the machine learning model captures complex nonlinear behaviors. This composite structure achieves superior model accuracy without excessive complexity, resolving the contradiction between simple control structure and reliable model accuracy.
2Measurement precision
If feedback control is increased to compensate for model inaccuracies, then tracking precision is maintained, but movements become stiff and injury risk increases
Solution Approach 1:
The hybrid model performs preliminary compensation for friction and nonlinear effects before feedback control is applied. By accurately predicting and compensating for friction torques and residual nonlinearities in advance, the system reduces the magnitude of corrective feedback needed during operation. This preliminary action maintains tracking precision while reducing stiff movements and injury risk associated with aggressive feedback control.
Solution Approach 2:
The patent converts the harmful effect of friction (which causes model inaccuracies and requires aggressive feedback) into a benefit by explicitly modeling it. The friction model transforms the previously harmful unmodeled effect into a useful predictive component, allowing the system to compensate for friction proactively. This conversion reduces the need for high-gain feedback control, maintaining precision while reducing injury risk from stiff movements.
3Reliability
If a friction model is added to account for dynamic effects, then model accuracy improves, but device complexity increases
Solution Approach 1:
The patent segments the modeling task into distinct components: rigid body dynamics, friction effects, and residual nonlinearities. Each segment is handled by a specialized sub-model with appropriate complexity. The friction model is segmented as a separate module with specific parameters for Coulomb and viscous friction. This segmentation improves model accuracy for each physical effect while keeping individual model components manageable in complexity.
Solution Approach 2:
The friction model introduces dynamic behavior to the system by accounting for velocity-dependent friction effects. Rather than using a static or overly complex model, the friction model dynamically adapts to changing operating conditions through velocity-based friction calculations. This dynamic approach improves model accuracy across varying speeds while maintaining reasonable structural complexity.
4Measurement precision
If machine learning model is used to capture residual torques, then tracking precision is improved, but computational requirements increase
Solution Approach 1:
The machine learning model is applied partially, focusing only on capturing residual nonlinearities that are not accounted for by the rigid body and friction models. Rather than using machine learning for the entire modeling task, the approach applies it selectively to the residual component. This partial application achieves improved tracking precision for complex nonlinear effects while limiting computational energy consumption to only where most needed.
Data Source
AI summary
A method for determining a torque for an operation of a robot that includes multiple components, using a model of the robot. Two components of the robot being movable using an actuator relative to one another at a connection point. The model maps parameters for the connection points onto a torque for the respective actuator. The method includes: providing values for the parameters including: a position or an angle a velocity, and an acceleration of the respective actuator; determining a value for the rigid body partial torque, the friction partial torque, and the machine learning partial torque, based on the values of the parameters and the model; determining a value of the torque of the respective actuator based on the values of the partial torques, and providing the value of the torque for the respective actuator for use during the operation, control or regulating of the robot.