SLAM

Novel transformer-enhanced physics-informed neural network (TE-PINN) achieving 36.8% error reduction in quaternion-based attitude estimation from IMU data. Integrates multi-head attention mechanisms with physics-based constraints (quaternion kinematics, rigid body dynamics) for robust real-time orientation estimation in high-noise, dynamic conditions.