🤖 Why Are Jeff Bezos and OpenAI Investing Millions in "Brains" for Robots?
Imagine a robot capable of performing any task, from folding laundry to assembling cars in a factory, just as ChatGPT generates text in any format. This is the vision behind the startup Physical Intelligence, which recently raised $400 million in funding from Amazon founder Jeff Bezos, OpenAI, and venture capital firms.
🖐 A New Approach
Typically, robots are designed for specific tasks, with each assistant requiring unique software. Physical Intelligence offers a fundamentally different approach. The company’s founder, Karol Hausman, calls it a "unified general brain" for robots. The company compares its model, π0, to GPT-1, which laid the foundation for the widespread adoption of large language models.
To create such a "brain," the development not only requires advanced AI algorithms and next-generation computer vision systems but also a massive dataset of how machines interact with the real world.
🤔 The Trainers' Race
Recently, the Chinese company Astribot demonstrated its assistant brewing coffee using π0. However, Physical Intelligence is far from the only player in this field.
💩 Robots using HIL-SERL software, developed at Berkeley, can learn to assemble computers or IKEA furniture in just a few hours and even play Jenga with a whip (yes, without losing!)
💩 Mobile ALOHA from Stanford has learned to call elevators, tidy up rooms, and cook shrimp. Initially, a human controls the robot manually, showing it how to perform a specific task. Afterward, the robot can replicate the task independently, even under changing conditions.
💩 Google DeepMind's AI agent RoboCat adapts to solving new tasks in just 100 iterations. It learns from its own mistakes and studies 3D simulations and videos of other robots at work.
💩 The startup Skild AI is making robots "curious." When a machine encounters an unfamiliar task, it attempts to complete it as many times as possible to gather maximum data for training.
More on the topic:
➡️ How Optimus Works: Elon Musk's Robot Assistant
➡️ Forbes Predicts Mass Adoption of General-Purpose Robots by 2030
