about use beta to fix Q

mbrossar / ai-imu-dr

AI-IMU Dead-Reckoning

MIT License

788 stars 222 forks source link

about use beta to fix Q #60

Closed kahowang closed 2 years ago

kahowang commented 3 years ago

Hi, thank you for your really greatful work. I am learing about AI-IMU,when i see the
alpha = self.factor_process_covariance(torch.ones(1).double()) beta = 10**(self.tanh(alpha))

i dont understand why to use beta to fix the Q? ,could you please show us more detailed steps. thanks very much.

hqtang-git commented 3 years ago

We can't clearly know the true value of Q, so zoom the approximation of Q by learning approrach, just for my own understanding.