Imagine that you are sweeping your house, cooking food, or watering the garden, and someone is paying you money for these tasks. It’s hard to believe, but a major bet on the future has made this possible. This is part of a new global campaign to train ‘humanoid robots’ (human-like machines). As of today, video clips of your household chores have become the most valuable data for those ‘android butlers’ (robotic assistants) who will take responsibility for your offices and kitchens in the coming times. Why are human videos necessary for machines? For decades, robots have been taught through remote control or virtual simulation (software). But in simulation, a robot cannot understand how much force is needed to pick up a glass. Shocking facts emerged from NVIDIA’s report When 20,000 hours of ‘first-person video’ (videos shot from a human’s perspective) were included in the robot’s training, their success rate increased by more than 50%. Now these robots are able to better perform complex tasks like folding T-shirts, sorting playing cards, and opening bottle caps. Billions of hours of data needed for robot training Human videos are needed to teach robots tasks. This is being called ‘egocentric data’ or ‘human data’. For this, startups worldwide are hiring contract workers. They wear cameras on their heads and perform everyday tasks. California-based company Micro1 has approximately 4,000 robotics trainers in 71 countries. These people are sending more than 1.6 lakh hours of video data every month. The company’s VP Aryan Sadeghi says this is just the beginning. Robots will need billions of hours of data for complete training. Fastest growth of data collection industry in Asia The market for data collection and labeling in the field of robotics is growing rapidly. According to market research firms, this industry is growing at an average annual rate of 30%. By 2030, it will be a market worth ~92,600 crores. The highest growth is being seen in Asia. According to companies like Objectways, American companies are paying 3 times more for data from American homes compared to countries like India. Objectways founder Ravi Rajalingam says, ‘It is more accurate to give the robot data from the place where it is to be deployed.’ Serious questions related to safety in household work Even though robots are working with 99.9% accuracy in factories, their success rate in the uncertain environment of homes is only between 70-80%. According to Alexander Verl, Chairman of the ‘International Federation of Robotics’, this is low for commercial use. The biggest danger is safety. Rajalingam warns, ‘If the robot cannot differentiate between a doll and a child, the consequences can be fatal. That’s why testing of robots with children is not happening, but data collection with pets has already begun.’
