Optimization of Gradient Descent Parameters in Attitude Estimation Algorithms

Area of Science:

• Aerospace Engineering
• Control Systems
• Robotics

Background:

• Attitude estimation is crucial for modern systems, relying on sensor data.
• Gradient descent is a recent optimal attitude estimation method.
• Parameter tuning for gradient descent is often overlooked.

Purpose of the Study:

• To investigate the impact of gradient descent parameters on attitude estimation quality.
• To propose a novel figure of merit and termination criterion for accuracy.
• To provide guidelines for optimal parameter selection for high accuracy and efficiency.

Main Methods:

• Analyzing step size, max iterations, initial quaternion, and propagation methods.
• Developing a new accuracy metric and termination criterion.
• Verifying proposed guidelines through simulations and experimental satellite ADCS model.

Main Results:

• Parameter choices significantly affect estimation quality in both noiseless and noisy data.
• The novel metrics effectively define and assess algorithm accuracy.
• Optimal parameter selection leads to higher accuracy with fewer iterations.

Conclusions:

• Gradient descent parameter tuning is critical for optimal attitude estimation.
• The proposed method and guidelines enhance accuracy and efficiency.
• The approach is suitable for low-power applications due to automatic iteration adjustment.