Working with both hands
Egocentric video captures tasks as they happen, including hand movements, objects and interactions.
Physical AI data from the real world
Capture the way people move, interact with objects and work in physical environments. 2une collects and annotates real world data for robots and other physical AI systems, from egocentric video to synchronized multimodal capture.
Samples
See the work as it happens
Physical AI needs more than images of objects. It needs to understand actions, movement, interaction and context. Our data captures real tasks from the person’s point of view, with the annotations and synchronized streams needed to train and evaluate physical AI systems.
Understanding the interaction
Hand and object annotations show what a person is doing, what they are interacting with and how the interaction changes over time.
Capturing the full scene
Synchronized camera and tracking data captures movement, objects and physical interactions together in real environments.
Real work, real environments
Data is collected in factories, kitchens, warehouses, healthcare settings and other environments where physical AI systems need to operate.
Capture
Capture what your model needs to learn
Different physical AI systems need different views of the world. We build the capture setup around the task.
Egocentric capture
First person video recorded from the operator’s point of view, using mono and stereo cameras across real working environments.
Multimodal capture
Multiple streams recorded together and synchronized throughout the session. Depending on the task, this can include motion capture, tactile sensing, stereo and depth cameras, wrist cameras and chest cameras.
Custom environments
When the right data does not exist in the right setting, we build the setting for it. We create controlled environments for specific tasks and scenarios, including kitchens, fulfilment spaces and clinical environments.
Pipeline
From capture to training ready data
The quality of physical AI data depends on what happens after recording.
Capture
We record real tasks using the capture setup required for the use case. Video, depth, motion, tactile and other streams can be collected together.
Quality control
Every stream is checked for synchronization, calibration and coverage. Sessions that do not meet the requirements are flagged and recaptured.
Annotation
We label actions, objects and interactions such as reaching, grasping and placing. Automated annotation handles scale, while human reviewers resolve ambiguity and edge cases.
Delivery
The final dataset is structured around your requirements, including aligned streams, annotations, metadata and calibration files.
Scale
Built for the Real world
Physical world data requires people, equipment and places to collect it. 2une brings these together through a trained workforce, dedicated facilities and access to real world environments across India.
10,000+
Trained experts
5
Factories
250+
Direct Partners
300+
Multistorey Building Infrastructure
Tell us what your model needs to learn
Give us the task, environment and data requirements. We will work with you to design the capture programme.