How do you balance model accuracy with inference speed for real-time robotic operations at Berkshire Grey?
