Teach your robot a new task.

How do you train a robot?

Start with one clear task. Collect examples, train a model and test whether the robot also performs the action under varying conditions.

Data collection

Good training data shows how your task is performed. Organise recording, quality and storage before you start collecting at scale.

AGIBOT software for training and simulation

AGIBOT develops software for data collection, model training, simulation and transferring models to robots. Which combination fits depends on your task and implementation.

What do you need to get started?

A robot alone is not enough to train a new task. Also prepare the workstation, software, data and people.

AGIBOT Genie Studio: from demonstration to robot skill

Discover how to organise a training programme: from the first task and demonstrations to evaluating a model and testing on the robot. AGIBOT's development environment; Humanoid-direct supports you with set-up and training.

AGIBOT Genie Sim: practise, measure and improve virtually

Build a virtual task environment, collect synthetic examples and compare robot models before trialling an application at the real workstation. AGIBOT's simulation platform: practise virtually before a practical test.

01

The task determines the training

We define the actions, objects, variation and required performance. We then choose which existing skills can be used and which training is still needed.

02

Collecting and preparing data

Depending on the task, we use demonstrations, teleoperation, sensor data or simulation. We check the quality, organise the examples and record which situations the dataset covers.

03

Training the robot

We use the collected data to train the required skill. We keep track of which data, configuration and model version were used, so that results remain comparable.

04

Testing with new situations

We also test on objects and conditions outside the training examples. This shows what works reliably, when a human needs to intervene and what additional training is needed.