Action Intelligence Launches Chorus, Promising 40x More Robot Training Data

Action Intelligence (行元智能) announced Chorus (众行 in Chinese) on October 27. It is a cross-embodiment data engine that turns video of human actions into pretraining data for robots of different designs. The claimed result is a 40x larger training set. The video comes from five types of wearable and handheld devices, and the main goal is to raise the reuse rate of human data.

Why human video has become robotics' currency

Language models had the internet. Robots don't. Every hour of real-world object manipulation has to be either recorded through teleoperation, where an operator in a headset or with controllers drives a robot arm, or generated in simulation, which transfers poorly to the physical world. The first route is expensive and slow: it requires robots, test sites and trained operators. The second suffers from the gap between simulator and reality.

Human video works differently. People pick up cups, open drawers and fold clothes every day anyway. If you film this with cheap wearable cameras, data piles up orders of magnitude faster. But there is a catch: a human has a hand with five fingers and a particular kinematic structure, while a robot may have a two-finger gripper, an anthropomorphic hand or a manipulator with a different number of degrees of freedom. This is the gap between "bodies," known as the embodiment gap. Chorus closes it at the data level: the source video is converted into a form suited to a specific platform.

Five device types in one pipeline

The announcement's strength is that Chorus is not tied to a single capture method. The company names the following sources:

• Smart glasses. They give a first-person view, as close as possible to what a robot's head "sees."
• Chest-mounted camera. It captures the work area in front of the body and the hands in frame without interfering with movement.
• Wrist-mounted camera. It shows a close-up of the grasp, which helps in training precise contact actions.
• UMI devices. Handheld grippers with a camera, modeled on the Universal Manipulation Interface. The human's motion is recorded in a way that makes it easier to transfer to a manipulator.
• Smart headband. A lightweight wearable sensor for long filming shifts.

Each device distorts the picture in its own way. Glasses see the whole scene, but hands can be occluded. A wrist camera shows contact accurately but loses context. A single engine that accepts such a heterogeneous stream spares the customer from building a separate pipeline for each sensor.

What the 40x gain promises

The number 40 in the release headline refers to the volume of training data available for pretraining after human videos have been run through Chorus. The logic is simple. If the same recording can be used for several types of robots, rather than only the platform it was filmed for, the total usable volume multiplies. Pretraining of robotics foundation models has long been constrained precisely by diversity and scale: without tens of thousands of hours of varied tasks, scenes and objects, models generalize poorly.

The key question calls for caution: how well the converted data performs on real robots. This is where cross-embodiment approaches are usually tested, by comparing task success rates for models with and without added human data. There are no independent measurements for Chorus yet, so it will have to be judged by future demonstrations and open benchmarks.

Who can use it in practice

For developers of humanoids and manipulators, Chorus promises to change the cost structure: part of the teleoperation budget could be replaced by cheap filming of people. Teams that already have their own fleets of wearable cameras get a chance to reuse their accumulated archives. Integrators and factories where operators perform standard assembly and warehouse tasks can record real staff work and turn it into training material.

Three things are worth watching: which robots are actually supported, what format the engine outputs data in, and whether Action Intelligence publishes comparisons against baseline methods. The answers will determine whether Chorus becomes a working layer of infrastructure or remains a polished presentation.