Training data for humanoid robots from your production
Humanoid robots are meant to work where people currently pick parts, supply materials or operate machines. Whether this works reliably depends less on the hardware than on the data they are trained with. We help companies translate their process knowledge into this kind of data.
ITCINITY does not build robots. For humanoid robotics, we capture, protect and structure the data, working with robotics partners.
Where humanoid robotics stands today
Several manufacturers show humanoid robots that carry boxes, sort parts or load machines. In industry, most of them run as pilot projects, not in continuous operation. Cycle time, reliability across entire shifts and the cost per working hour are still open questions.
Humanoids are interesting where the environment is built for people: shelves, stairs, workbenches, machines with control panels. For many tasks, a robotic arm or a mobile robot is the safer and cheaper choice today. We check this for each use case before any data is captured.
Why videos alone are not enough
Recognising a component, gripping it securely and fitting it correctly comes naturally to experienced employees. A robot has to coordinate perception, movement and force to do this, with changing part positions, different materials or a workpiece that has slipped.
Videos of successful runs show only part of this. Training also requires motion data, contact forces, the result of each attempt and recordings of how people correct errors. The goal is for the robot to perform a task safely and to the required quality again and again under changing conditions.
How training data for robots is created
Usually several methods are combined. Which ones depends on the task, the target robot and the required accuracy.
Learning from demonstration
Skilled workers carry out the process, and cameras and sensors record every step.
Teleoperation
A person controls the robot remotely. This produces image, motion and state data that fit the robot directly.
Instrumented handheld grippers
Portable grippers with cameras record handling tasks at the workstation without a robot having to be there yet.
Simulation and synthetic data
Digital environments add further variants and deliberately generated faults to real recordings.
Learning from operation
Failed attempts and human corrections, captured in a controlled way, show where the model is still weak.
Whatever works in simulation has to be checked again on the real robot.
From work process to trained robot
In five steps, your employees’ knowledge becomes a dataset that robotics partners can train with.
Select the application
Review the work process, quality requirements and cost-effectiveness. Clarify whether a humanoid robot, a robotic arm or a mobile system fits best.
Capture work processes
Record typical work steps, their variants and the corrections made by experienced employees with suitable sensors.
Prepare and protect the data
Synchronise recordings, assess them and provide them in a structured form. Define access rights, confidentiality and usage rights from the start.
Adapt and test models
Train with the robotics partner and measure success by part quality, cycle time and the human support still required.
Improve in a targeted way
Capture further data where the test shows weaknesses. Check every updated model again before it is used.
What this is suitable for
Suitable are recurring handling tasks in which people currently spend a lot of time gripping, carrying and placing.
- Removing, sorting and placing components in the correct position
- Supplying materials and components for assembly
- Loading and unloading machines and test stations
- Recurring handling and packaging tasks
A deployment can reduce physically demanding routine work, make processes more consistent and give more flexibility with changing variants. The experience of long-serving employees is also preserved in structured form. Whether a deployment pays off depends on the individual case, because besides the hardware, integration, training, support, maintenance and possible downtime all cost money.
Data protection, works council and safety
Anyone who films employees at work or records their movements is processing personal data. Purpose, legal basis and retention period have to be defined beforehand, employees are informed, and depending on the scope, a data protection impact assessment is required. Because cameras and sensors can record behaviour and performance, the works council generally has to be involved.
Recordings from production often contain trade secrets. We can process them locally on the AI Cube and pass on to robotics partners only the data needed for the agreed purpose. Who may use which data is set out in the contract.
A robot in operation also needs a risk assessment and a safety assessment of the overall system. From 20 January 2027, the new EU Machinery Regulation applies to this. This assessment and practical tests are part of every implementation.
Who does what
ITCINITY
- Analysing work processes and assessing use cases
- Planning the capture and carrying it out with suitable sensors
- Storing data locally, preparing it and providing it in a structured form
- Organising data protection, access rights and usage rights
- Connection to your IT and production systems
Robotics partners
- Robot hardware and grippers
- Training and adapting the robot models
- Integration into the cell or line
- Safety assessment and commissioning
- Maintenance of the robot system
Frequently asked questions about training data for robots
What is training data for robots?
Recordings of how a task is carried out: camera images, movements, gripping and contact forces, machine states and the result of each attempt. From this, a robot model learns to carry out the task itself.
How is training data for humanoid robots captured?
Through demonstration with cameras and sensors, through teleoperation of a robot, with instrumented handheld grippers, supplemented by simulation and data from ongoing operation. Usually several methods are combined.
Is a humanoid robot already worthwhile for our company today?
For most tasks, a robotic arm or a mobile robot is more cost-effective today. Humanoids are mainly interesting where the environment is built for people and converting it would be expensive. Companies that capture process data in a structured way now can use it for both routes.
Are we allowed to record employees for robot training?
Yes, with a clear purpose, a legal basis, information for employees and a limited retention period. Depending on the scope, a data protection impact assessment is required, and the works council generally has to be involved. We plan the capture accordingly.
Does ITCINITY build robots?
No. We analyse processes, capture and protect the data and prepare it. Robots, training of the models and commissioning are handled by specialised robotics partners.
Does our production data stay with us?
We define that. Recordings can be processed locally on the AI Cube, and robotics partners only receive the agreed data for the agreed purpose.
Assessing humanoid robots for your production
Talk to us about your use case and the prerequisites for a first pilot. The initial consultation is free of charge.
