The Astribot S1 humanoid robot can make coffee autonomously with training from Physical Intelligence’s buzzy AI model that helps robots see, understand, and act.
In new demo footage, Astribot, developed by China-based Stardust Intelligence, uses the π₀ (Pi Zero) model to follow high-level instructions, such as “make coffee,” by breaking them into smaller, actionable steps like adding water, measuring coffee grounds, and pressing buttons. Its vision allows it to identify objects such as mugs and coffee makers even when they are misplaced or in unexpected locations. This adaptability enhances its performance in dynamic environments.
The π₀ model features a multi-layered training approach:
- General Knowledge: The model’s training includes diverse datasets from multiple robots, providing Astribot with foundational skills such as grasping, pouring, and pressing buttons.
- Specific Fine-Tuning: π₀ is further fine-tuned with coffee-making data, enabling Astribot to execute tasks like brewing coffee or frothing milk efficiently and accurately. This fine-tuning ensures it knows the necessary steps and tools right from the start.
Simple tasks like brewing drip coffee take approximately 2–3 minutes, while more complex processes, such as finding misplaced items or creating specialty drinks, may require additional time.
Physical Intelligence, a Silcon Valley startup founded less than a year ago, is already valued at $2.4 billion. The model’s ability to generalize across tasks, adapt to new environments, and recover from errors has attracted significant interest from the robotics and AI community.
The π₀ model enables Astribot to handle tasks flexibly by combining knowledge from diverse datasets and adapting to new situations. This approach represents a departure from traditional robotics, which often rely on rigid programming and struggle in unpredictable scenarios.
Announced at the 2024 World Robot Conference, the Astribot is designed for tasks such as folding clothes, cooking, and cleaning. It can even participate in activities like practicing martial arts and shooting basketballs. The robot’s AI adapts and improves over time, enhancing its efficiency.