The long tail lives outside the lab.
Cluttered kitchens, worn tools, bad lighting, a hundred ways to fold a shirt. Models only generalise after they've seen the mess.
Robots learn from how people actually work.
We record real physical work in homes, factories, construction sites, mines and farms, then turn it into synced, labelled data that robot learning teams can train on.
Session in progress
AccelGyroHand pose
The problem
Robots stall where the lab ends.
Real workplaces are hard to get into.
Factories, mines and construction sites need access, safety clearance and trained people on the ground before a single frame is recorded.
Footage isn't a dataset.
Video only becomes training data once it's synced with motion, checked for quality, segmented into actions and packaged for your stack.
What we capture
Every channel a policy needs, recorded together.
CH 1
Egocentric video
First-person footage from head-mounted cameras, hands and tools always in frame.
CH 2
IMU
Acceleration and rotation, time-synced to every video frame.
CH 3
Motion capture
Body and hand pose for tasks that need precise kinematics.
CH 4
Teleoperation
Operator-driven robot demonstrations, logged as actions and states.
Work with us
Start with data on the shelf, or tell us the task.
Off-the-shelf dataset
Over 1,000 hours of egocentric video with synced IMU, recorded across everyday household tasks. Ready to license today.
1,010 hrs / ~2 TB / 15-min clipsCustom collection
You define the task and the success criteria. We recruit and train operators, capture at our facility, and deliver in batches.
Scoped per programOn-site industrial capture
Recording inside working factories and sites, with access, safety and operator logistics handled by us.
Factory / construction / mining / agricultureOperations
Built on people and places, not just devices.
~100trained capture operators working in shifts
100+factory sites across Odisha, Rajasthan and Uttarakhand
6environment types, from homes to mines
1,000+hours of off-the-shelf data available now
Delivery
A dedicated residential capture facility in Bhubaneswar, a managed device fleet and a quality review on every batch. Data ships in the format your pipeline already reads.
RLDSLeRobotMCAPRaw + metadata
FAQ
Common questions.
What does Plexus Robotics do?
We collect real-world data of people doing physical tasks and prepare it for robot learning. That covers the recording itself, the people and sites behind it, and the processing that turns raw footage into a usable dataset.
Who is it for?
Robotics companies, humanoid teams and AI labs training manipulation or embodied models that need task-specific, real-world demonstrations.
Where is the data captured?
At our residential capture facility in Bhubaneswar, and on-site at factories and working environments across Odisha, Rajasthan and Uttarakhand.
What formats do you deliver in?
RLDS, LeRobot and MCAP, or raw files with metadata if you run your own pipeline.
Is Plexus Robotics the same as Grid?
Yes. Plexus Robotics Private Limited is the company; Grid is the name our data operations have traded under.
How do we get started?
Email us with the task, environment and volume you have in mind. We usually start with a sample or a small pilot before scaling up.
Tell us what your robot needs to learn.
For dataset licensing, custom collection and pilots. We reply within two working days.
hello@plexusrobotics.in