Methods to detect failure of optimization

[Jason]

I am wondering if there is a way to detect the failure of bundle adjustment? I am using GTSAM to optimize the pose estimation from visual-inertial odometry. I found the optimization randomly fails even running on the same dataset, same initialization (except slightly different values of bias of IMU). When the optimization fails, the estimated velocity is much larger than ground truth. But the value of velocity is definitely not a good indicator of failure since the range is application dependent. Could you give me a hint?

[Frank Dellaert]

Bias can have a large effect. You can detect by checking if residual changed, and number of iterations.... Closing as not a bug, really.

[Frank Dellaert]

Not a bug, gave some guidance.