Accelerate Your Data Flywheel and Transform Teleoperation into Intelligent Automation
- louisegoldsmith1
- Mar 2
- 3 min read
We’re excited to share our AMAS AI Pipeline Tutorial - a step-by-step guide designed to help you turn teleoperated robotic tasks into scalable, AI-powered automation workflows.
Built on our AMAS (Autonomous Manipulation & AI Software) platform and deployed through Extend Robotics, this tutorial walks you through the entire process of collecting data, training models, and deploying autonomous capability, without writing a single line of code.
If you’ve ever wondered how to move from manual robot control to intelligent automation, this is your roadmap.
Watch our tutorial video now:
Step 1: Record Tasks Using “Data Collection”
It starts with demonstration. Simply operate the robot to perform your desired task using the Data Collection feature. As you control the robot, AMAS automatically records movements, gestures, actions, environmental context.
There’s no coding. No scripting. No complex configuration. The robot learns directly from real-world demonstrations, capturing how tasks are actually performed, not how they’re theoretically programmed.
Step 2: Visualise Tasks as Episodes in the User Portal
Once recorded, AMAS organises each task attempt into structured Episodes within the User Portal.
These Episodes provide a clear breakdown of each demonstration, visual replay of task execution, insight into how the robot interpreted each action.
This structured view makes it easy to understand what was captured and prepares your data for refinement.
Step 3: Data Annotation – Turning Demonstration into Intelligence
Raw data becomes powerful when it’s refined.
Using AMAS’s integrated data annotation tools, operators can label key task stages, identify important interaction points, add contextual metadata.
This transforms demonstrations into high-quality, human-verified intelligence, significantly improving model training outcomes.
Step 4: Curate Your Dataset
Not every demonstration will be perfect, and that’s expected. AMAS allows you to review each Episode and flag any that should not be used for training. Removing incomplete or low-quality recordings ensures stronger model performance, greater consistency, higher automation reliability.
Data integrity is critical, and AMAS gives you full control over what your model learns from.
Step 5: Train a New Model
With curated Episodes selected, you’re ready to train. AMAS offers multiple baseline AI model algorithms, giving flexibility depending on task complexity, environment variability, application requirements.
The system automatically processes your high-quality demonstrations and builds a model capable of independently replicating the task. This is where teleoperation transforms into scalable automation.
Step 6: Deploy the Model to Cortex
Once training is complete, deployment is seamless. Your refined AI model can be instantly deployed to Cortex, Extend Robotics’ powerful robot control platform.
This creates a smooth transition from learning to real-world execution
Step 7: Switch to AI Mode
With your model deployed, you can activate AI Mode. Now the robot infers the correct actions, makes real-time decisions, controls timing and movement autonomously, executes the task independently.
Operators are freed from repetitive manual control yet can intervene at any time to resume full teleoperation. This hybrid control approach ensures both efficiency and safety.
The AMAS Data Flywheel
The real power of AMAS isn’t just training once. It’s continuous improvement.
Collect data. Train. Deploy. Collect more data.Improve the existing model.
Each cycle strengthens performance, expands capability, and moves your robot closer to fully optimised automation.
Many organisations struggle with the jump from teleoperation to autonomy. The gap often lies in data quality, model training complexity, and deployment friction.
The AMAS AI Pipeline removes those barriers.
Find out more about how Extend Robotics can transform your automation by contacting us today.
