Real-World Robot Training Data

Bringing Robot Training Into the Real World

Robots are becoming increasingly capable, but building systems that can operate reliably in the real world requires experience beyond controlled labs and demonstrations. Homes, hotels, factories, warehouses, construction sites, and other workplaces introduce different objects, layouts, workflows, and situations that are difficult to fully recreate in a training environment.

This project aims to create more opportunities to deploy robots into real businesses where they can perform practical tasks while generating valuable training data. By connecting robotics developers with businesses across different industries, we can create environments where robots can learn from real-world experience while exploring useful applications for the organizations hosting them.

Why Real-World Robot Data Matters

Physical AI models need large amounts of diverse data to learn how robots should perceive, move, manipulate objects, and respond to changing situations. While simulated and controlled environments are valuable for development, real businesses introduce the variation that robots ultimately need to understand, from unfamiliar objects and layouts to people, equipment, and constantly changing surroundings.

Deploying robots into working environments creates an opportunity to collect this experience while performing meaningful tasks. Each deployment can expose robotic systems to new situations and provide developers with data that can be used to train, evaluate, and improve models for increasingly capable real-world robots.

Tutor Intelligence built DF1: 100 robots running real-world tasks simultaneously.

Real-World Environments

Useful robot training can happen anywhere physical work is being done. Hotels, warehouses, machine shops, construction sites, restaurants, retail spaces, and other businesses each provide different environments, objects, equipment, and workflows for robots to encounter.

The project will explore opportunities across industries, working with businesses to identify practical tasks that could support both robotics development and their own operations. This could range from moving and organizing materials to handling objects, assisting workers, restocking supplies, or completing other repetitive physical tasks.

Creating Value Through Deployment

Real-world robot training should create value for the businesses participating in it. Rather than treating a workplace simply as a testing environment, deployments can focus on tasks that businesses already need performed and explore where emerging robotic systems could make those operations more efficient.

Businesses can participate in focused pilots, provide feedback on how robots perform, and help identify the problems that are most valuable to solve. Robotics developers gain access to real operating conditions and meaningful tasks, while participating businesses get an opportunity to explore new automation technologies as they develop.

Isaac the robot folding laundry inside a laundry room.

Working With Robotics Companies

Robotics and physical AI companies need access to a wide range of environments to test their systems and collect data from real tasks. The project can help connect these companies with businesses that have relevant workflows, equipment, objects, and operating conditions for their technology.

Each collaboration can be structured around the needs of the technology being developed. Companies may bring their own robots, models, data requirements, or research objectives, while deployments provide an opportunity to evaluate those systems in real operating environments and gather experience that can contribute to further development.

Building a Real-World Deployment Network

The long-term goal is to build a growing network of businesses interested in participating in robotics development. Different industries and locations can provide new environments, tasks, objects, and challenges, creating opportunities for robotics companies to deploy systems across a much broader range of real-world conditions.

As the network grows, new robotics technologies can be matched with businesses that have relevant use cases and are interested in exploring automation. This can make it easier to move promising systems out of controlled development environments and into the places where they will ultimately be used.

Physical Intelligence unveils π0, a general-purpose robot foundation model.

For Robotics and Physical AI Companies

We are looking to work with robotics and physical AI companies that need access to real-world environments for training, testing, and deployment. This can include companies developing foundation models, autonomous systems, manipulation technologies, mobile robots, humanoids, robotic arms, or other emerging physical AI systems.

Collaborations can take many forms depending on the technology and development goals. We can help identify relevant businesses and environments, explore potential tasks, coordinate deployments, and create opportunities to collect real-world experience and training data.

Industries and Businesses

We are looking for businesses interested in exploring how emerging robotics technologies could be used in their operations. Potential environments include hospitality, manufacturing, logistics, construction, property operations, retail, food service, agriculture, and other industries where physical work is performed every day.

Participating businesses do not need to have an existing robotics program or a specific automation project in mind. We are interested in learning how work is currently performed, identifying repetitive or challenging physical tasks, and exploring where real-world robot deployments could create value.