Leader-follower teleoperation
Leader-follower teleoperation is a method for controlling a robot, and recording demonstrations, in which a person moves a leader device and a follower robot copies its motion in real time. The leader is usually a second arm with the same joint layout as the follower, so each joint maps directly. It is the standard way to collect training data on low-cost arms such as the SO-101 and on bimanual systems such as ALOHA.
In this article
Building visual or physical AI?
Let's talk.
What is leader-follower teleoperation?
In a leader-follower setup, the operator never touches the robot doing the work. They move the leader arm by hand, sensors in its joints read the positions, and software sends those positions to the follower arm as commands. Because the two arms share the same joint layout, the mapping is direct, with no need to translate hand motions into robot coordinates.
This direct mapping is why leader-follower rigs produce clean demonstrations for imitation learning. The recorded actions are exactly the joint targets the follower received, the follower's measured positions become the robot's state, and cameras capture what the robot saw. Low-cost systems such as ALOHA and the SO-101, along with rigs like GELLO that add a leader arm to an existing industrial arm, have made the approach widespread.
Key takeaways
- A person moves a leader arm, and a follower arm mirrors its joint positions in real time.
- Matching joint layouts make the mapping direct, which produces clean demonstration data.
- It is the standard recording method for low-cost arms in the LeRobot ecosystem.
How it works
Both arms are calibrated so the same physical pose produces the same joint readings. During recording, software reads the leader's joint positions many times per second, sends them to the follower as targets, and logs the targets as actions alongside the follower's measured state and camera frames. In LeRobot, each recorded attempt becomes an episode in a LeRobot dataset.
Why it matters
Leader-follower teleoperation matters because demonstration quality limits imitation learning, and this method makes high-quality demonstrations cheap. It also has limits worth knowing: operators can only demonstrate what the leader's design allows, there is usually no force feedback, and fatigue over long sessions shows up as hesitations and failed attempts in the data, which curation needs to catch.
Frequently asked questions
How is leader-follower teleoperation different from VR teleoperation?
Leader-follower rigs map joint positions directly from a matching arm. VR teleoperation tracks the operator's hands or controllers and converts those poses into robot commands, which works across different robot designs but adds a translation step.
What data does leader-follower teleoperation record?
Typically the commanded joint positions as actions, the follower's measured joint positions as state, and synchronized camera frames, all timestamped and grouped into episodes.
Related terms
Jesse Mostipak
Director of Growth
Jesse Mostipak is the Director of Growth at Voxel51, where the work is helping humans find and trust what the brand knows, and teaching the Google knowledge graph and the LLMs answering on their behalf to do the same. That question, how knowledge gets built inside a system, is one Jesse has been chasing for years. Earlier versions of it ran through a New York City high school science classroom, data science and machine learning, and developer relations at Kaggle, Posit (formerly RStudio), and Baseten. The answer doesn't change much depending on whether the learner is a teenager, a software engineer, or a knowledge graph. Jesse holds a Master's in Education from CUNY Hunter College. LinkedIn
See all articles by Jesse Mostipak
Last updated October 8, 2026
Building visual or physical AI?
Let's talk.
© 2026 Voxel51 All Rights Reserved