Research on Gradient-Descent Extended Kalman Attitude Estimation Method for Low-Cost MARG

Area of Science:

• Robotics
• Sensor Fusion
• Navigation Systems

Background:

• Gradient descent methods suffer from weak dynamic performance and susceptibility to sensor interference in attitude and heading reference systems.
• Nonlinear Kalman Filter methods present computational complexity challenges for low-cost MARG sensor systems.

Purpose of the Study:

• To propose an Extended Kalman Filter (EKF) method that combines a two-stage gradient descent algorithm with EKF (GDEKF) for low-cost MARG sensor systems.
• To enhance the dynamic performance, anti-interference capabilities, and accuracy of attitude estimation.

Main Methods:

• A two-stage gradient descent algorithm corrects attitude angles using accelerometer and magnetometer data.
• The corrected attitude quaternion and gyroscope measurements form the observation vector for EKF.
• EKF corrects gyroscope-derived attitude and bias, improving observational stability.

Main Results:

• The GDEKF method demonstrates superior anti-interference and dynamic performance compared to the standard gradient descent method.
• The proposed GDEKF achieves better measurement accuracy than the traditional Extended Kalman Filter.
• Simulations confirmed the effectiveness of the GDEKF for MARG sensor systems.

Conclusions:

• The GDEKF offers a robust and accurate solution for attitude and heading estimation in low-cost MARG sensor systems.
• This approach effectively mitigates sensor noise and improves system stability and responsiveness.