VADER: Visual Affordance Detection and Error Recovery for Multi Robot Human Collaboration