MoRPI-PINN: a physics-informed framework for mobile robot pure inertial navigation.

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Fusion

Background:

• Accurate mobile robot navigation is essential for full autonomy, especially when GPS or cameras are unavailable.
• Inertial sensors alone suffer from drift due to noise and errors, limiting pure inertial navigation.
• Maneuvering robots in a snake-like motion can enhance the inertial signal-to-noise ratio to mitigate drift.

Purpose of the Study:

• To propose MoRPI-PINN, a physics-informed neural network framework for drift mitigation in inertial-based mobile robot navigation.
• To investigate and improve the pure inertial navigation solution for mobile robots.
• To demonstrate the effectiveness of physics-informed neural networks in enhancing navigation accuracy.

Main Methods:

• Developed MoRPI-PINN, a physics-informed neural network framework tailored for inertial navigation.
• Embedded physical laws and constraints within the neural network's training process.
• Validated the approach using real-world experimental data with unseen trajectories.

Main Results:

• MoRPI-PINN achieved over 80% accuracy improvement compared to baseline approaches for navigation on unseen trajectories.
• The framework provides an accurate and improved navigation solution by leveraging physical principles.
• Demonstrated the feasibility of implementing MoRPI-PINN on edge devices for practical mobile robot applications.

Conclusions:

• MoRPI-PINN offers a robust solution for accurate mobile robot navigation in GPS-denied environments.
• The physics-informed approach significantly enhances the reliability and precision of inertial navigation.
• This lightweight framework is suitable for widespread adoption in various mobile robot applications.