log in
3 types of datacollection. Here you can explain what these datasets contain or just something funny, you choose. At least make the text about this long, because that looks good for my lay-out ok thank you.
A human controls a leader arm while a follower robot mirrors the motion in real time. This setup enables precise recording of trajectories, joint positions, and manipulation tasks while keeping data collection intuitive for operators.
Robots can also be controlled using handheld controllers and camera systems. By recording control commands together with visual observations, we can create datasets that connect human intent with robotic actions.
Using head-mounted devices, we capture actions directly from a human perspective. This approach allows us to record how people interact with objects and environments while preserving natural movement and decision-making processes.
We’re Diego and Sanne, two builders based in Mexico exploring efficient methods for collecting high-quality robotic training data.
We started by designing and printing robotic arms, but quickly realized that the biggest challenge in robotics isn’t hardware, it’s data.
Many robot learning systems depend on large amounts of high-quality data, but collecting that data is often expensive and difficult. We’re exploring simpler and more scalable ways to capture robotic training data and better understand what makes useful datasets for robot learning.
We’re currently building and testing affordable data collection systems using sensors, cameras, and robots.
Our focus is on creating reliable data pipelines, capturing human demonstrations, and exploring how different collection methods affect dataset quality.
Right now, we’re focused on collecting data. Our long-term goal is to use that data for future robot learning experiments and to better understand the workflows and tools that make robotic learning possible.