Skip to content
Bud's Wiki

AI Sapiens Open Source — ROBOTIS Docs Directory

TL;DR

Section titled “TL;DR”

The AI Sapiens open-source docs page is ROBOTIS’s canonical directory of the public repositories that make up the AI Sapiens K1 humanoid stack — ROS 2 platform code (ai_sapiens), simulation + RL (cyclo_lab), robot-description files (URDF + MuJoCo, USD forthcoming), motion generation and retargeting (GEM-X, soma-retargeter, kimodo — each forked from an NVIDIA / NV Labs upstream), and shared actuator-control layers (DynamixelSDK, dynamixel_hardware_interface, cyclo_intelligence). Hardware BOM, STEP files, and tutorial videos are announced as planned releases. Software is Apache-2.0, hardware CERN-OHL-P-2.0, docs CC BY 4.0.

Key claims

Section titled “Key claims”
  • The K1 open-source stack is organized around two entry points: ROBOTIS-GIT/ai_sapiens for the ROS 2 software stack and ROBOTIS-GIT/cyclo_lab for simulation-based policy training and testing [§AI Sapiens Repositories].
  • Robot-description assets are published in two formats — URDF at ai_sapiens_description/urdf/k1_rev1 and MuJoCo at ai_sapiens_description/mujoco/k1; USD is announced but not yet public [§Simulation Models].
  • The motion generation and retargeting layer consists of three ROBOTIS-maintained forks of NVIDIA / NV Labs repositories: ROBOTIS-GIT/GEM-X (from NVlabs/GEM-X, video-to-motion generation), ROBOTIS-GIT/soma-retargeter (from NVIDIA/soma-retargeter), and ROBOTIS-GIT/kimodo (from nv-tlabs/kimodo) [§Motion Generation and Retargeting].
  • Shared actuator-control and learning components — dynamixel_hardware_interface (ROS 2 hardware interface), DynamixelSDK, and cyclo_intelligence — back the wider AI Sapiens stack [§Shared ROBOTIS Software].
  • Licensing is split by domain: Apache-2.0 for software, CERN-OHL-P-2.0 for hardware, CC BY 4.0 for documentation, with per-repo LICENSE and third-party notices as the source of truth [§Open Source Licenses].
  • Planned but not-yet-released resources are called out explicitly: complete Hardware BoM, mechanical STEP files, and tutorial videos [§Planned Resources].

Method

Section titled “Method”

The page is a directory / index, not a technical paper — it enumerates repository URLs, filesystem paths inside those repos, and pointer relationships (fork lineage). The organizational structure is four tiers: (1) AI Sapiens Repositories — the K1-specific ROS 2 and RL entry points; (2) Simulation Models — the robot-description assets in URDF and MuJoCo formats; (3) Motion Generation and Retargeting — the K1-specific forks of upstream NVIDIA / NV Labs motion pipelines (GEM-X for video-to-motion, soma-retargeter for retargeting, kimodo for the retargeting-path composition); (4) Shared ROBOTIS Software — actuator SDKs and control interfaces that pre-date AI Sapiens but are consumed by it. The page then declares “planned resources” (BoM, STEP files, tutorial videos) as forward-looking, and terminates with the license split.

Results

Section titled “Results”

No benchmarks or model weights — this is an infrastructure directory. The concrete deliverables are: the public GitHub URLs, the fork-lineage attribution (three of the motion-side repos are declared forks of NVIDIA upstreams, which is significant for anyone tracing the provenance of the motion stack), and the license grid.

Why it’s interesting

Section titled “Why it’s interesting”

Complements Cyclo — ROBOTIS Open-Source Modular Physical AI Framework filed a day earlier: Cyclo is the framework umbrella, this page is the directory — the specific set of concrete repositories that make up the K1 stack, plus the pieces (ai_sapiens, robot-description files, GEM-X / soma-retargeter / kimodo) that don’t sit under the Cyclo umbrella but are part of the same open-source rollout. The explicit fork lineage — three ROBOTIS-maintained forks of NVIDIA / NV Labs repos on the motion side — is the most interesting factual claim: it makes concrete the “hardware vendor takes NVIDIA-lab research code and hardens / integrates for one specific platform” pattern that the Open foundation-model releases concept has been tracking. Contrasts with the vertically-integrated Chinese embodied stacks like Qwen-Robot Suite (Qwen-Robot Suite launch — Qwen-RobotManip, Qwen-RobotNav, Qwen-RobotWorld + RobotClaw framework (Alibaba Qwen)) and RynnBrain 1.1 (RynnBrain 1.1 — Open Embodied Foundation Models (Alibaba DAMO Academy)) where the model lab owns everything end-to-end — here a hardware vendor is stitching upstream research forks together and publishing the seams.

See also

Section titled “See also”