Everyday tasks such as clearing the dining table and loading the dishwasher present a significant dexterity challenge for home robots that can require a lot of effort. training data and capital.
The new startup says it spent less than two years and a fraction of the cost figuring it out.
On Thursday, Sunday Robotics came out of stealth mode to demonstrate Memo, a fully autonomous system. home robot on wheels that can do household chores.
A video posted on X by company co-founder Tony Zhao shows Memo moving from the dining room to the kitchen to clear dishes from the table and load them into the dishwasher. The company said Memo performed the task autonomously.
Another feat involved Memo lifting two wine glasses with one hand, which are known to be very fragile. The robot also folded socks and loaded the coffee machine.
Sunday Robotics, also known as Sunday, was founded in April 2024 by Zhao and Cheng Chi, both of whom have backgrounds in robotics.
On Sunday, more than 500 data collectors across the US are training startup Memo's robot. Best regards, Sunday
“Today we introduce a step forward in robotic artificial intelligence,” Zhao said in the X-post. The co-founder added that Memo hasn't broken a single glass in more than 20 live demo sessions.
Teaching a robot to interact with common household objects, some of which may be fragile, is a critical test of dexterity in the world of robotics.
First, copy human handhaving thousands of sensory receptors is a complex engineering challenge in itself. Tesla CEO Elon Musk This was stated in the company's most recent earnings report in October.
The information used to train robots is also a major bottleneck.
Many companies have turned to teleoperations, in which a person controls the robot using joysticks or various controllers, to train robots. Other companies are experimenting with synthetic data and modeling.
None of these common methods are used on Sunday. Instead, according to the startup's co-founder, the company created a patented glove that mimics the shape of Memo's Lego-like hands.
The person wears gloves and performs certain tasks that provide Memo with data, such as the force used to lift an object.
Zhao said the method is a more efficient and cost-effective way to train robots. In Post X, he said the glove provides “two orders of magnitude better capital efficiency than teleoperation ($200 vs. $20,000).”
Zhao added that it is also scalable because data can be collected anywhere without the need to carry Memo around. The startup employs more than 500 data collectors across the US who provide training data for Memo.
“If the only thing we can rely on in robotics is teleoperation, then it would certainly take decades to collect that amount of training data,” Zhao told TBPN.
