How AI Companies Hiring Small Town Workers To Train Robots

As artificial intelligence continues to dominate our digital lives, its next frontier lies beyond the screen in understanding how we move in the real world. Across the globe, a rising number of trainers are guiding AI to grasp human motion, helping it step out of computers and into our homes, workplaces, and factories, reports the Los Angeles Times.

A report by Nilesh Christopher highlighted how AI companies are recording visuals which involve basic human movements to train AI to do the same.

He gives the example of Naveen Kumar, who works for a startup which needs such data. Kumar’s job is to record footage of folding towels.

“He mounts a GoPro camera to his forehead and follows a regimented list of hand movements to capture exact point-of-view footage of how a human folds.”

“If it takes more than a minute or he misses any steps, he has to start over,” Christopher writes.

Kumar works for Objectways, a data-labelling firm. It recently provided its US client with 200 videos demonstrating how to fold towels. The company employs over 2,000 people, around half focus on tagging sensor data for self-driving cars and robotics, while the rest contribute to generative AI projects.

“Sometimes we have to delete nearly 150 or 200 videos because of silly errors in how we’re folding or placing items,” said Kumar.

Kumar and his team, based in the town of Karur, located roughly 483 km away from Bengaluru, are an unexpected group of instructors for the upcoming generation of AI-driven robots, writes Christopher.

In the United States, companies such as Tesla, Boston Dynamics, and Nvidia are at the forefront of developing the next wave of robotic technology. Tesla has begun showcasing its Optimus robots at various company events. Google continues to advance its own AI-driven robotics models, while OpenAI is expanding its efforts in the same field.

Nvidia estimates that the global market for humanoid robots could be worth around $38 billion (about Rs 3,36,921 crore) within the next ten years.

Ali Ansari, founder of the San Francisco-based company Micro1, explained that teleoperation is becoming a key focus in the collection of robotics data. 

In this approach, humans use controllers to guide robots through tasks such as picking up a cup or preparing tea. The AI system is then trained by analysing videos of both successful and unsuccessful attempts, allowing it to learn how to perform these actions independently.

As demand rises, an increasing number of establishments, sometimes referred to as “arm farms”, are emerging, according to Mohammad Musa, the founder of the California-based data annotation company Deepen AI.

“Today, a mix of real and synthetic data is being used, gathered from human demonstrations, teleoperation sessions and staged environments,” he said. “Much of this work still occurs outside the West, but automation and simulation are reducing that dependency over time.”