The power of the machine reports.
Edit: sia
Another player who has accelerated the arrival of the home robot "chatgpt" moment, has appeared.The bot startup remained shrouded in mystery until Meta's chief AI scientist, Yann Lecun, took to the social platform to raise news of its hiring and announced that one of the founders was his colleague at NYU, Lerrel Pinto.
At present, there is almost zero news on the Internet about the robot company called Fauna Robotics, and there is not even a company's official website, and there is only a brief introduction to the official account of the X platform, and they are "creating a world where human and robot species communities coexist, collaborate and prosper together." Still, we can get some clues about the company from the word "fauna" and the background of one of the key figures, Lerel Pinto.
To date, Roomba and other robotic vacuum cleaners are the only truly successful commercial home robots. What makes them so successful is that they don't need to interact with humans and their environment – in fact, they just learn to avoid them. Fauna stands for Zoology, and it seems that the start-up wants to describe and document the history of a new robotic species community, that is, robots that can perform daily household chores in different home environments and live together with humans. This is not different from the ambitions of Tesla Optimus Prime, Figure AI, Digit to 1 X and other humanoid robot companies, all of whom are accelerating the arrival of the "ChatGPT" moment for home robots.
In addition, from the research background of one of the company's founders, Lerrel Pinto, the company's mission should also be related to universal home robots. Lerrel Pinto, a 31-year-old Ph.D. from CMU's Robotics Institute, has been engaged in research on adapting robots to the human living world, and was named one of MIT Technology Review's 35 Under 35 innovators in 2023. Currently, Lerrel Pinto is an assistant professor in the Department of Computer Science at New York University Courant and a member of the New York University Laboratory for Computational Intelligence, Vision, and Robotics (CILVR, founded by Yann Lecun).
According to LinkedIn, Lerrel Pinto joined Fauna a month ago (in February, which could also be the company's founding). Coincidentally, in January, NYU and Meta jointly released a new open-knowledge robotics framework, OK-Robot. This is a new general-purpose robot that integrates large language models (LLMS), visual language models (VLMs) and robots to complete the tasks of picking, moving, and dropping objects in a realistic home environment without any additional training (which is often time-consuming and expensive).
You just need to say to the robot, bring the cornflakes on the table to the bedside table, and the robot can find the cornflakes by itself, and plan the route and corresponding actions to complete the task smoothly. On top of that, it can also help you organize your things or take out the trash.
Running tests in 10 real-world home environments showed that the OK-Robot had a 58 success rate for pick-and-place tasks5%, and if the home environment is relatively clean and tidy, the success rate can be increased to 82%.
In 10 home environments in New York City, OK-Robot attempted 171 pick-up tasks. **Links:
One of the biggest challenges for robots to do their daily chores in a completely unfamiliar environment is that it takes a lot of data to train them, and we can scrape massive amounts of data from the internet to train LLMS, but it's hard for robots to do that – training the robot's data requires collecting real-world data, which makes it more difficult to build and scale training datasets.
Pinto and his team came up with a simple, easy-to-replicate way to expand the robot's training dataset — connect an iPhone to a common garbage picker and use it to record themselves grabbing objects in their home, and the iPhone's built-in hardware (such as lidar, motion sensors, and gyroscopes) record important data for tasks such as motion, depth, and rotation — important for training the robot to replicate its movements on its own. After accumulating a certain amount of data, they are used to combine it with efficient learning algorithms to train models and teach robots how to do household chores.
In addition to OK-Robot, Pinto and his team have also open-sourced Dobb-E — also using data collected from real homes to train models, and the robot has learned household chores like turning on the air fryer, closing the door, or straightening the cushion. A 30-day test in 10 New York homes found that the robot completed 109 chores with an overall success rate of 81% and mastered the new skill in just 20 minutes.
In addition to Lerrel Pinto, the other founder of Fauna Robotics is Rob CPCHRAN. According to LinkedIn data, he joined the company two days ago and is also the company's CEO. However, a month ago, it was revealed that the former Goldman Sachs managing director quietly left after three and a half years at Goldman Sachs and started his own robotics company. 
Prior to joining Goldman Sachs, Cochran spent most of his career at the "Internet Giant" (FAANG). Prior to joining neural interface startup Ctrl-Labs, he worked as a product manager on the Amazon Cloud for three years. Upon joining CTRL-Labs, he quickly rose through the ranks from Senior Product Manager to Head of Product Management. In 2019, Ctrl-Labs was acquired by Facebook, and Cochran continued to work at Facebook for eight months before moving to Goldman Sachs. Rob Cochran was previously Chief Product Officer in Goldman Sachs' engineering division. Lerrel Pinto has previously said in an interview with ** that he hopes to see robots in the home not only do vacuum work, but also become a more integral part of our lives, such as doing housework, taking care of the elderly, etc., and can be there at any time when we need it. He also hopes that one day it will be able to learn new tasks without having to show the robot more examples. In the eyes of some roboticists, Pinto and his team's research suggests that the future of home robots is really here. It's no longer just some crazy dreams. "We're seeing a machine learning revolution that allows the newly created models that are now being created not only to work in the lab, but also to work well in the open world. It's very useful information to see that they really work in a real physical environment. According to Matthias Minderer, senior computer vision research scientist at Google's Deepmind. Pinto wants to accelerate the arrival of the moment of large languages for robots and help usher in a new era of AI. "Once people start believing that domestic robots are possible, there will be more results in this area. "There is an argument that action is what motivates humans to have brains. "Evolution is first and foremost about keeping us alive and finding food." But he argues, "The ultimate goal of intelligence is to act and change the world, and I think the only people who can do that are physical creatures, such as robots." Reference link.