# ROBOTIS Teaches Humanoid Robot K-POP Dance from Video > Source: > Published: 2026-06-12 12:58:29.568837+00:00 # ROBOTIS Teaches Humanoid Robot K-POP Dance from Video Reporting by Interesting Engineering describes a demonstration in which **ROBOTIS** used its open-source **AI Sapiens** platform to teach a humanoid robot a complex full-body routine, the **CORTIS REDRED Challenge**, using only smartphone video. Reporting by Interesting Engineering and blogger Mike Kalil says the pipeline combined video-based motion capture, motion retargeting, simulation-based reinforcement learning, and Sim2Real transfer to move policies from a digital twin to the physical robot. Mike Kalil reports the robot is **1.3 meters** tall, weighs **34 kg**, has **23 degrees of freedom**, and runs on an 8-core ARM CPU with a Mali GPU and an NPU; he also notes ROBOTIS has not confirmed pricing or a release date. Editorial analysis: Industry practitioners should view this as another example of lowering the barrier to humanoid motion learning through cheaper capture and open-source toolchains. ### What happened Reporting by Interesting Engineering describes that **ROBOTIS** demonstrated its open-source **AI Sapiens** humanoid platform learning the **CORTIS REDRED Challenge** dance using only smartphone video rather than professional motion-capture systems. Reporting by Interesting Engineering and Mike Kalil says the demonstration combined video-based motion capture, motion retargeting, simulation-based reinforcement learning, and Sim2Real transfer to move learned behaviors from a simulated digital twin to a physical robot. Mike Kalil reports the robot measures **1.3 meters** in height, weighs **34 kg**, has **23 degrees of freedom**, and runs onboard inference on an **8-core ARM** CPU with a **Mali GPU** and a dedicated **NPU**; Kalil also notes ROBOTIS has not confirmed a release date or pricing, though Korean reports cited by Kalil place potential price below **$10,000**. ### Technical details Editorial analysis - technical context: The public descriptions place the pipeline stages that matter to practitioners into four parts: smartphone video capture; pose extraction and motion retargeting to a simulated humanoid; reinforcement learning in simulation to optimize balance and timing; and Sim2Real transfer for deployment on hardware. Reporting highlights the use of **DYNAMIXEL-Q** actuators in the physical robot, which are presented as part of the hardware stack supporting the transfer of learned policies to actuated joints. ### Context and significance Editorial analysis: For robotics researchers and engineers, the two notable trends in this demonstration are the use of commodity capture (smartphone video) to generate training targets and the publication of an open-source physical-AI framework. Comparable public demonstrations increasingly emphasize lowering setup cost for imitation learning; that pattern accelerates experimentation because teams can iterate on motion datasets without access to studio-grade mocap. The reported hardware specs make **AI Sapiens** roughly comparable on paper to other recent lower-cost humanoids, though public reporting does not provide benchmarking data on stability, repeatability, or sample efficiency. ### What to watch Editorial analysis: Observers should look for a public code or dataset release, technical documentation of the retargeting and RL reward design, and controlled evaluations comparing sim-to-real robustness to existing platforms. Pricing and availability statements from ROBOTIS would clarify whether this demonstration translates into a broadly accessible research platform. Additionally, independent tests measuring repeatability across runs and robustness to variations in phone-camera quality will be key indicators of practical utility. ## Scoring Rationale A notable open-source humanoid robotics demo showing a smartphone-video-to-physical-motion pipeline (imitation learning, Sim2Real) - relevant to robotics ML practitioners. Sector-specific with limited independent verification, placing it at the top of the solid range. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)