Meet the gig workers helping train robots to do our dishes

Mohamad Dunggio, a 23-year-old resident of Gorontalo on Sulawesi in Indonesia, has taken to recording himself doing household chores in order to earn extra cash. For four hours straight, he donned a mobile phone on his head and captured his actions, leaving him with wrinkled fingers but content with the additional income. He is one of thousands of gig workers worldwide contributing to the burgeoning robotics industry by generating egocentric training data, which involves recording tasks from the perspective of the person performing them, typically with a head or chest-mounted camera.

These workers, spanning factories in South Asia, house cleaners in the US, and delivery drivers, are paid between $3 and $10 an hour to perform tasks such as dishwashing, cleaning, cooking, making the bed, ironing, folding clothes, and sweeping. Professor Dana Kulić, a robotics expert at Monash University, emphasizes that robots must learn to observe and react to the diverse and complex environments within homes to reliably perform domestic tasks.

By using egocentric data, similar to the training of large language models like ChatGPT, robots can better understand and adapt to the unique characteristics of each individual's living space.

While some contributors, like Dunggio, find the gig work rewarding and not exploitative, others have expressed concerns about privacy, understanding the purpose of data collection, and fees for withdrawing earnings. Robotics companies such as Appen, based in Australia, collect and annotate datasets, including egocentric data, from a global pool of contributors.

Despite facing criticism over pay and conditions, Appen maintains a code of ethics and fair-pay standards, striving to ensure payments are always above the minimum wage in the respective country. However, challenges remain in achieving reliable, cost-effective robotics for everyday household tasks.