Energy-Based Inverse Dynamics Model for Multi-Actuator Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Formulating an accurate model of inverse dynamics for mechanical systems like robotic manipulators is challenging due to parameter uncertainty and the difficulty in describing complex dynamics such as motor friction and joint elasticity, which limits their control accuracy.
Innovation Solution
An energy-based inverse dynamics model using Gaussian Processes Regression (GPR) is trained with a full prior and posterior covariance matrix to capture correlations between torques of different actuators, allowing for more accurate torque mapping and control command determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a data-driven model is used to approximate complex non-linear dynamics, then the model can capture complex dynamics, but it suffers from low data efficiency and poor generalization properties
Solution Approach 1:
The patent transforms the inverse dynamics model from a direct torque prediction approach to an energy-based approach by changing the fundamental parameters being modeled. Instead of modeling torques directly, the model learns kinetic energy and potential energy parameters, which then derive torques through physical relationships. This parameter transformation improves generalization and reduces data requirements while maintaining accuracy in capturing complex non-linear dynamics.
Solution Approach 2:
The patent replaces the traditional mechanical system modeling approach with a machine learning-based energy model. Instead of using physics-based mechanical models with uncertain parameters, the system uses a learned energy model that captures complex dynamics without requiring explicit physical parameter knowledge, thereby improving reliability while reducing the need for extensive parameter identification.
2Measurement precision
If an energy-based inverse dynamics model with full covariance matrix is used, then correlations between torques are captured accurately, but the model complexity increases
Solution Approach 1:
The patent merges the modeling of multiple torque components into a unified energy-based framework. Instead of modeling each torque independently, the system models kinetic and potential energy that inherently capture correlations between all torques. The full covariance matrix in the Gaussian Process formulation further merges uncertainty information across all torque dimensions, achieving accurate correlation capture while maintaining computational tractability through the energy-based formulation.
Data Source
AI summary
The present disclosure discloses a system and a method for controlling a mechanical system having different actuators and multiple degrees of freedom to track a reference trajectory for performing a task. The system comprises a memory configured to store an energy based inverse dynamics model trained with machine learning to map states of the different actuators to corresponding torques for the different actuators, wherein the energy based inverse dynamics model is configured to model energy of the mechanical system with a Gaussian Processes Regression (GPR) process having a matrix capturing correlations between the torques of the different actuators. The system further comprises a processor configured to process the states of the different actuators with the energy based inverse dynamics model to produce values of the torques for the different actuators of the mechanical system, and control the mechanical system based on the produced values of the torques.