Solanki Dharak Deepak

I am currently using smolVLA for SO101 arm. I collected about 200 episodes of a total recording time of 1 hour 5 min . After training it for 30k steps and deployed , I have observed that the model works decent but it does not give 100% accuracy. I have used 2 cameras and my task is pick and place. Is there any specific thing I am missing in my setup ? Like the position of the cameras or the variability in the dataset. I have used teleoperatio to collect these many episodes. I would be very grateful for any sights you could share