[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74145":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":16,"starSnapshotCount":16,"syncStatus":29,"lastSyncTime":30,"discoverSource":31},74145,"kimodo","nv-tlabs\u002Fkimodo","nv-tlabs","Official implementation of Kimodo, a kinematic motion diffusion model for high-quality human(oid) motion generation.","",null,"Python",2552,276,27,4,0,26,81,253,78,29.33,"Apache License 2.0",false,"main",[],"2026-06-12 02:03:22","\u003Cp align=\"center\">\n  \u003Cimg src=\".\u002Fassets\u002Fbanner.png\" alt=\"Banner\" width=\"100%\">\n  \u003Ca href=\"LICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache%202.0-76B900.svg\" alt=\"License\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FProject-Page-blue\" alt=\"Project Page\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Findex.html\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-online-green.svg\" alt=\"Documentation\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n## Overview\n\nKimodo is a **ki**nematic **mo**tion **d**iffusi**o**n model trained on a large-scale (700 hours) commercially-friendly optical motion capture dataset. The model generates high-quality 3D human and robot motions, and is controlled through text prompts and an extensive set of constraints such as full-body pose keyframes, end-effector positions\u002Frotations, 2D paths, and 2D waypoints. Full details of the model architecture and training are available in the [technical report](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fassets\u002Fkimodo_tech_report.pdf).\n\nThis repository provides:\n- **Inference**: code and CLI to generate motions on both human and robot skeletons\n- **Interactive Demo**: easily author motions with a timeline interface of text prompts and kinematic controls\n- **Benchmark**: [test cases](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fnvidia\u002FKimodo-Motion-Gen-Benchmark) and evaluation code built on the [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) dataset to evaluate motion generation models based on text and constraint-following abilities\n- **Annotations**: fine-grained temporal text descriptions created for the Kimodo project are included in the [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) dataset. For more information on these labels, see our separate [Hugging Face repo](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fnvidia\u002FSEED-Timeline-Annotations).\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fteaser.gif\" width=\"1280\">\n\u003C\u002Fdiv>\n\n## News\n\nSee the [full changelog](CHANGELOG.md) for a detailed list of all changes.\n\n- **[2026-05-03]** _FIX_: fixed a bug causing incorrect calculation of averaged metrics for constraint test cases in the benchmark\n- **[2026-04-24]** _NEW_: improved multi-prompt generation and better support for small VRAM GPUs via `TEXT_ENCODER_DEVICE=cpu` env var\n- **[2026-04-10]** Released the [Kimodo Motion Generation Benchmark](#kimodo-motion-generation-benchmark) alongside new v1.1 Kimodo-SOMA models\n- **[2026-03-19]** **Breaking:** Model inputs\u002Foutputs now use the SOMA 77-joint skeleton (`somaskel77`).\n- **[2026-03-16]** Initial open-source release of Kimodo with five model variants (SOMA, G1, SMPL-X), CLI, interactive demo, and timeline annotations for BONES-SEED.\n\n\n## Kimodo Models\n\nSeveral variations of Kimodo are available trained on various skeletons and datasets. All models support text-to-motion and kinematic controls.\n\n> Note: models will be downloaded automatically when attempting to generate from the CLI or Interactive Demo, so there is no need to download them manually\n\n| Model | Skeleton | Training Data | Release Date | Hugging Face | License |\n|:-------|:-------------|:------:|:------:|:-------------:|:-------------:|\n| **Kimodo-SOMA-RP-v1.1** | [SOMA](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FSOMA-X) | [Bones Rigplay 1](https:\u002F\u002Fbones.studio\u002Fdatasets#rp01) | April 10, 2026 | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-RP-v1.1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-SOMA-SEED-v1.1** | [SOMA](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FSOMA-X) | [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) | April 10, 2026  | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-SEED-v1.1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-SOMA-RP-v1** | [SOMA](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FSOMA-X) | [Bones Rigplay 1](https:\u002F\u002Fbones.studio\u002Fdatasets#rp01) | March 16, 2026 | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-RP-v1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-G1-RP-v1** | [Unitree G1](https:\u002F\u002Fgithub.com\u002Funitreerobotics\u002Funitree_mujoco\u002Ftree\u002Fmain\u002Funitree_robots\u002Fg1) | [Bones Rigplay 1](https:\u002F\u002Fbones.studio\u002Fdatasets#rp01) | March 16, 2026  | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-G1-RP-v1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-SOMA-SEED-v1** | [SOMA](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FSOMA-X) | [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) | March 16, 2026  | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-SEED-v1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-G1-SEED-v1** | [Unitree G1](https:\u002F\u002Fgithub.com\u002Funitreerobotics\u002Funitree_mujoco\u002Ftree\u002Fmain\u002Funitree_robots\u002Fg1) | [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) | March 16, 2026  | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-G1-SEED-v1) | [NVIDIA Open Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-open-model-license\u002F) |\n| **Kimodo-SMPLX-RP-v1** | [SMPL-X](https:\u002F\u002Fgithub.com\u002Fvchoutas\u002Fsmplx) | [Bones Rigplay 1](https:\u002F\u002Fbones.studio\u002Fdatasets#rp01) | March 16, 2026  | [Link](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SMPLX-RP-v1) | [NVIDIA R&D Model](https:\u002F\u002Fwww.nvidia.com\u002Fen-us\u002Fagreements\u002Fenterprise-software\u002Fnvidia-internal-scientific-research-and-development-model-license\u002F) |\n\nBy default, we recommend using the models trained on the full Bones Rigplay 1 dataset (700 hours of mocap) for your motion generation needs.\nThe models trained on BONES-SEED use 288 hours of [publicly available mocap data](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) so are less capable, but are useful for comparing to other models trained on BONES-SEED. To easily compare motion generation models to Kimodo, check out our [Motion Generation Benchmark](#kimodo-motion-generation-benchmark).\n\n### Changes in v1.1\nThe latest v1.1 Kimodo-SOMA models were released primarily for compatibility with our new [Motion Generation Benchmark](#kimodo-motion-generation-benchmark), but also contain minor quality improvements over v1. For details on these improvements, please see the Hugging Face pages for [Kimodo-SOMA-RP-v1.1](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-RP-v1.1#changes-in-v11) and [Kimodo-SOMA-SEED-v1.1](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FKimodo-SOMA-SEED-v1.1#changes-in-v11).\n\n## Getting Started\n\nPlease see the full documentation for detailed installation instructions, how to use the CLI and Interactive Demo, and other practical tips for generating motions with Kimodo:\n\n**[Full Documentation](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs)**\n- [Quick Start Guide](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fgetting_started\u002Fquick_start.html)\n- [Installation Instructions](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fgetting_started\u002Finstallation.html)\n- [Interactive Motion Authoring Demo](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Finteractive_demo\u002Findex.html)\n- [Command-Line Interface](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fuser_guide\u002Fcli.html)\n- [Benchmark Instructions](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fbenchmark\u002Fintroduction.html)\n- [API Reference](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fapi_reference\u002Findex.html)\n\n**Before getting started** with motion generation, please review the [best practices](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fkey_concepts\u002Flimitations.html) and be aware of [model limitations](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fkey_concepts\u002Flimitations.html#limitations).\n\n\nSome notes on installation environment:\n- Kimodo requires ~17GB of VRAM to generate locally entirely on GPU, primarily due to the text embedding model. If you have a smaller card, set `TEXT_ENCODER_DEVICE=cpu` when running Kimodo commands to force text encoding to the CPU. This is slightly slower but reduces VRAM usage to \u003C3 GB.\n- The model has been most extensively tested on GeForce RTX 3090, GeForce RTX 4090, and NVIDIA A100 GPUs, but should work on other recent cards with sufficient VRAM\n- This repo was developed on Linux, though Windows should work especially if using Docker\n\n## Interactive Motion Authoring Demo\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fdemo_screenshot.png\" width=\"1000\">\n\u003C\u002Fdiv>\n\n\u003C\u002Fbr>\n\n**[Demo Documentation and Tutorial](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Finteractive_demo\u002Findex.html)**\n\nThe web-based interactive demo provides an intuitive interface for generating motions with any of the Kimodo model variations. After installation, the demo can be launched with the `kimodo_demo` command. It runs locally on http:\u002F\u002F127.0.0.1:7860. Open this URL in your browser to access the interface (or use port forwarding if set up on a server).\n\n### Demo Features\n- **Multiple Characters**: Supports generating with the SOMA, G1, and SMPL-X versions of Kimodo\n- **Text Prompts**: Enter one or more natural language descriptions of desired motions on the timeline\n- **Timeline Editor**: Add and edit keyframes and constrained intervals on multiple constraint tracks\n- **Constraint Types**:\n  - Full-Body: Complete joint position constraints at specific frames\n  - 2D Root: Define waypoints or full paths to follow on the ground plane\n  - End-Effectors: Control hands and feet positions\u002Frotations\n- **Constraint Editing**: Editing mode allows for re-posing of constraints or adjusting waypoints\n- **3D Visualization**: Real-time rendering of generated motions with skeleton and skinned mesh options\n- **Playback Controls**: Preview generated motions with adjustable playback speed\n- **Multiple Samples**: Generate and compare multiple motion variations\n- **Examples**: Load pre-existing examples to better understand Kimodo's capabilities\n- **Export**: Save constraints and generated motions for later use\n\n## Command-Line Interface\n\n**[CLI Documentation and Examples](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fuser_guide\u002Fcli.html)**\n\nMotions can also be generated directly from the command line with the `kimodo_gen` command or by running `python -m kimodo.scripts.generate` directly.\n\n**Key Arguments:**\n- `prompt`: A single text description or sequence of texts for the desired motion (required)\n- `--model`: Which Kimodo model to use for generation\n- `--duration`: Motion duration in seconds\n- `--num_samples`: Number of motion variations to generate\n- `--constraints`: Constraint file to control the generated motion (e.g., saved from the web demo)\n- `--diffusion_steps`: Number of denoising steps\n- `--cfg_type` \u002F `--cfg_weight`: Classifier-free guidance (`nocfg`, `regular` with one weight, or `separated` with two weights for text vs. constraints); see the [CLI docs](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fuser_guide\u002Fcli.html#classifier-free-guidance-cfg)\n- `--no-postprocess`: Flag to disable foot skate and constraint cleanup post-processing\n- `--seed`: Random seed for reproducible results\n\nThe script supports different output formats depending on which skeleton is used. By default, a custom NPZ format is saved that is compatible with the web demo.\nFor Kimodo-G1 models, the motion can be saved in the standard MuJoCo qpos CSV format.\nFor Kimodo-SMPLX, motion can be saved in the standard AMASS npz format for compability with existing pipelines.\n\n### Default NPZ Output Format\nGenerated motions are saved as NPZ files containing:\n- `posed_joints`: Global joint positions `[T, J, 3]`\n- `global_rot_mats`: Global joint rotation matrices `[T, J, 3, 3]`\n- `local_rot_mats`: Local (parent-relative) joint rotation matrices `[T, J, 3, 3]`\n- `foot_contacts`: Foot contact labels [left heel, left toe, right heel, right toes] `[T, 4]`\n- `smooth_root_pos`: Smoothed root representations outputted from the model `[T, 3]`\n- `root_positions`: The (non-smoothed) trajectory of the actual root joint (e.g., pelvis) `[T, 3]`\n- `global_root_heading`: The heading direction output from the model `[T, 2]`\n\n`T` the number of frames and `J` the number of joints.\n\n## Low-Level Python API\n\n**[Model API Documentation](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fapi_reference\u002Fmodel.html#kimodo.model.kimodo_model.Kimodo.__call__)**\n\nFor maximum flexibility, the low-level model inference API can be called directly, rather than going through our high-level CLI.\nThis allows for advanced model configuration including classifier-free guidance weights and parameters related to transitions in multi-prompt sequences.\n\n## Downstream Robotics Applications of Kimodo\n\n### Visualizing G1 Motions with MuJoCo\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fmujoco_result.gif\" width=\"800\">\n\u003C\u002Fdiv>\n\nAfter generating motions on the G1 robot skeleton and saving to the MuJoCo qpos CSV file format, they can be easily used and visualized within MuJoCo.\nA minimal visualization script is available with:\n```\npython -m kimodo.scripts.mujoco_load\n```\nMake sure to edit the script to correctly point to your CSV file and install Mujoco before running this.\n\n### Tracking Generated Motions with ProtoMotions\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fprotomotions_results.gif\" width=\"1280\">\n\u003C\u002Fdiv>\n\n[ProtoMotions](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FProtoMotions) is a GPU-accelerated simulation and learning framework for training physically simulated digital humans and humanoid robots. The Kimodo NPZ and CSV output formats are both compatible with ProtoMotions making it easy to train physics-based policies with generated motions from Kimodo. ProtoMotions supports outputs on both the SOMA skeleton and Unitree G1\n\nAfter generating motions with Kimodo, head over to the [ProtoMotions docs](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FProtoMotions?tab=readme-ov-file#-motion-authoring-with-kimodo) to see how to import them.\n\n### Retargeting Motions to Other Robots with GMR\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fgmr_results.gif\" width=\"1280\">\n\u003C\u002Fdiv>\n\nMotions generated by Kimodo-SMPLX can be retargeted to other robots using [General Motion Retargeting (GMR)](https:\u002F\u002Fgithub.com\u002FYanjieZe\u002FGMR).\nGMR supports the AMASS NPZ format out of the box, so simply generate motions with Kimodo and use `--output` to save; the AMASS NPZ is written to `stem_amass.npz` (single sample) or in the output folder (multiple samples). Then, use the [SMPL-X to Robot script](https:\u002F\u002Fgithub.com\u002FYanjieZe\u002FGMR?tab=readme-ov-file#retargeting-from-smpl-x-amass-omomo-to-robot) in GMR to retarget to any supported robot. For example:\n```\n# run within GMR codebase\npython scripts\u002Fsmplx_to_robot.py --smplx_file \u002Fpath\u002Fto\u002Fsaved\u002Famass_format.npz --robot booster_t1\n```\n\n### Combining Kimodo with GEAR-SONIC\n\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"assets\u002Fsonic_kimodo_demo.gif\" width=\"800\">\n\u003C\u002Fdiv>\n\nAs a proof of concept, we have also incorporated Kimodo into the [interactive GEAR-SONIC demo](https:\u002F\u002Fnvlabs.github.io\u002FGEAR-SONIC\u002Fdemo.html). In the demo, Kimodo can be used to generate a kinematic motion on the G1 robot skeleton, then GEAR-SONIC tracks the motion in simulation.\n\n## Kimodo Motion Generation Benchmark\n\n[**[Benchmark Documentation](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fbenchmark\u002Fintroduction.html)**]\n[**[Test Suite on Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fnvidia\u002FKimodo-Motion-Gen-Benchmark)**]\n\nAlongside the Kimodo models, we provide a benchmark designed to standardize evaluation for motion generation models with a comprehensive set of test cases. This includes:\n\n* **Evaluation Data**: A suite of test cases [available on Hugging Face](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fnvidia\u002FKimodo-Motion-Gen-Benchmark) is used in concert with the [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) dataset to construct the full benchmark. \n* **Diverse Test Cases**: Test cases cover a wide range of text-conditioned and constraint-conditioned motion generation.\n* **Evaluation Pipeline**: Code for the full evaluation pipeline including benchmark construction, motion generation, and evaluation.\n* **Metrics**: Several metrics to evaluate generated motions that cover motion quality, constraint following, and text alignment. Our [TMR-SOMA-RP-v1](https:\u002F\u002Fhuggingface.co\u002Fnvidia\u002FTMR-SOMA-RP-v1) model trained on all 700 hours of the Bones Rigplay dataset is a powerful embedding model to compute common metrics like R-precision and FID.\n\nTo facilitate future research, we [report benchmark results](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fdocs\u002Fbenchmark\u002Fresults.html) for Kimodo-SOMA-v1.1 models, which are reproducible and easily comparable to other methods trained on the BONES-SEED data. \n\n## Timeline Annotations for BONES-SEED\n\nAs detailed in the [tech report](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Fsil\u002Fprojects\u002Fkimodo\u002Fassets\u002Fkimodo_tech_report.pdf), Kimodo is trained using fine-grained temporal text annotations of mocap clips.\nWhile the full [Rigplay 1](https:\u002F\u002Fbones.studio\u002Fdatasets#rp01) dataset is proprietary, we have released the temporal segmentations for the public [BONES-SEED](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) subset.\nThese annotations are already included in the BONES-SEED dataset, but the standalone labels and additional information about them is [available on HuggingFace](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fnvidia\u002FSEED-Timeline-Annotations).\n\n\n## Related Humanoid Work at NVIDIA\nKimodo is part of a larger effort to enable humanoid motion data for robotics, physical AI, and other applications.\n\nCheck out these related works:\n* [SOMA Body Model](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FSOMA-X) - a unified parameteric human body model\n* [BONES-SEED Dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fbones-studio\u002Fseed) - a large scale human(oid) motion capture dataset in SOMA and G1 format\n* [ProtoMotions](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FProtoMotions) - simulation and learning framework for training physically simulated human(oid)s\n* [SOMA Retargeter](https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fsoma-retargeter) - SOMA to G1 retargeting tool\n* [GEM](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FGEM-X) - human motion reconstruction from video\n* [GEAR SONIC](https:\u002F\u002Fgithub.com\u002FNVlabs\u002FGR00T-WholeBodyControl) - humanoid behavior foundation model for physical robots\n\n## Citation\n\nIf you use this code in your research, please cite:\n\n```bibtex\n@article{Kimodo2026,\n  title={Kimodo: Scaling Controllable Human Motion Generation},\n  author={Rempe, Davis and Petrovich, Mathis and Yuan, Ye and Zhang, Haotian and Peng, Xue Bin and Jiang, Yifeng and Wang, Tingwu and Iqbal, Umar and Minor, David and de Ruyter, Michael and Li, Jiefeng and Tessler, Chen and Lim, Edy and Jeong, Eugene and Wu, Sam and Hassani, Ehsan and Huang, Michael and Yu, Jin-Bey and Chung, Chaeyeon and Song, Lina and Dionne, Olivier and Kautz, Jan and Yuen, Simon and Fidler, Sanja},\n  journal={arXiv:2603.15546},\n  year={2026}\n}\n```\n\n## License\n\nThis codebase is licensed under [Apache-2.0](LICENSE). Note that model checkpoints and data are licensed separately as indicated on the HuggingFace download pages.\n\nThis project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.\n\n## Acknowledgments\n\nThis project builds upon excellent open-source projects:\n- [Viser](https:\u002F\u002Fgithub.com\u002Fnerfstudio-project\u002Fviser) for 3D motion authoring demo\n- [LLM2Vec](https:\u002F\u002Fgithub.com\u002FMcGill-NLP\u002Fllm2vec) for text encoding\n\n## Contact\n\nFor questions or issues, please open an issue on this repository or reach out directly to the authors.\n\n---\n","Kimodo是一个用于生成高质量3D人体和机器人动作的运动扩散模型。该项目基于大规模（700小时）商用友好型光学动作捕捉数据集训练而成，能够通过文本提示及多种约束条件如全身姿态关键帧、末端执行器位置\u002F旋转、2D路径等来控制生成的动作。其核心功能包括生成高精度的人形与机械骨骼动画、提供交互式演示界面以便捷地创建动画序列，并且附带了基准测试用例与评估代码，便于开发者验证模型在遵循文本指令和物理约束方面的能力。适用于游戏开发、虚拟现实体验设计以及机器人动作规划等领域。",2,"2026-06-11 03:49:01","high_star"]