{"id":44124567,"url":"https://github.com/NVIDIA/Isaac-GR00T","last_synced_at":"2026-02-20T20:00:37.038Z","repository":{"id":283166409,"uuid":"946829547","full_name":"NVIDIA/Isaac-GR00T","owner":"NVIDIA","description":"NVIDIA Isaac GR00T N1.6 -  A Foundation Model for Generalist Robots.","archived":false,"fork":false,"pushed_at":"2026-02-14T00:25:35.000Z","size":55091,"stargazers_count":6203,"open_issues_count":211,"forks_count":1013,"subscribers_count":76,"default_branch":"main","last_synced_at":"2026-02-16T23:36:22.149Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"https://developer.nvidia.com/isaac/gr00t","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/NVIDIA.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-03-11T18:34:24.000Z","updated_at":"2026-02-16T20:42:07.000Z","dependencies_parsed_at":"2025-09-19T06:15:37.084Z","dependency_job_id":"e98e110d-a908-4296-8c7a-cf3f19e6316f","html_url":"https://github.com/NVIDIA/Isaac-GR00T","commit_stats":null,"previous_names":["nvidia/isaac-gr00t"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/NVIDIA/Isaac-GR00T","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2FIsaac-GR00T","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2FIsaac-GR00T/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2FIsaac-GR00T/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2FIsaac-GR00T/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/NVIDIA","download_url":"https://codeload.github.com/NVIDIA/Isaac-GR00T/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/NVIDIA%2FIsaac-GR00T/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29662556,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-20T19:49:36.704Z","status":"ssl_error","status_checked_at":"2026-02-20T19:44:05.372Z","response_time":59,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.6:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-02-08T21:01:11.713Z","updated_at":"2026-02-20T20:00:37.030Z","avatar_url":"https://github.com/NVIDIA.png","language":"Jupyter Notebook","funding_links":[],"categories":["related: Robotics \u0026 Foundation models","Foundation Models","Python","🔗 Useful Resources"],"sub_categories":["Medical \u0026 Assistive","Base VLA Models"],"readme":"\u003cdiv align=\"center\"\u003e\n\n  \u003cimg src=\"media/header_compress.png\" width=\"800\" alt=\"NVIDIA Isaac GR00T N1.6 Header\"\u003e\n\n  \u003c!-- --- --\u003e\n  \n  \u003cp style=\"font-size: 1.2em;\"\u003e\n    \u003ca href=\"https://developer.nvidia.com/isaac/gr00t\"\u003e\u003cstrong\u003eWebsite\u003c/strong\u003e\u003c/a\u003e | \n    \u003ca href=\"https://huggingface.co/nvidia/GR00T-N1.6-3B\"\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/a\u003e |\n    \u003ca href=\"https://huggingface.co/datasets/nvidia/PhysicalAI-Robotics-GR00T-X-Embodiment-Sim\"\u003e\u003cstrong\u003eDataset\u003c/strong\u003e\u003c/a\u003e |\n    \u003ca href=\"https://arxiv.org/abs/2503.14734\"\u003e\u003cstrong\u003ePaper\u003c/strong\u003e\u003c/a\u003e |\n    \u003ca href=\"https://research.nvidia.com/labs/gear/gr00t-n1_6/\"\u003e\u003cstrong\u003eResearch Blog\u003c/strong\u003e\u003c/a\u003e\n  \u003c/p\u003e\n\u003c/div\u003e\n\n## NVIDIA Isaac GR00T\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"media/stacked_demo.gif\" width=\"800\" alt=\"GR00T Demo\"\u003e\n\u003c/div\u003e\n\n\u003e We just released GR00T N1.6, an updated version of GR00T N1 with improved performance and new features. Check out the [release blog post](https://research.nvidia.com/labs/gear/gr00t-n1_6/) for more details.\n\n\u003e To use the older version, N1.5, please checkout the [n1.5-release](https://github.com/NVIDIA/Isaac-GR00T/tree/n1.5-release) branch.\n\nNVIDIA Isaac GR00T N1.6 is an open vision-language-action (VLA) model for generalized humanoid robot skills. This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments.\n\nGR00T N1.6 is trained on a diverse mixture of robot data including bimanual, semi-humanoid and an expansive humanoid dataset. It is adaptable through post-training for specific embodiments, tasks and environments.\n\nThe neural network architecture of GR00T N1.6 is a combination of vision-language foundation model and diffusion transformer head that denoises continuous actions. Here is a schematic diagram of the architecture:\n\n\u003cdiv align=\"center\"\u003e\n\u003cimg src=\"media/model-architecture.png\" width=\"800\" alt=\"model-architecture\"\u003e\n\u003c/div\u003e\n\nHere is the general procedure to use GR00T N1.6:\n\n1. We assume the user has already collected a dataset of robot demonstrations in the form of (video, state, action) triplets for a specific task. \n2. The user will first convert the demonstration data into the LeRobot compatible data schema (more info in [`getting_started/data_preparation.md`](getting_started/data_preparation.md)), which is compatible with the upstream [Huggingface LeRobot Dataset V2](https://github.com/huggingface/lerobot).\n3. Our repo provides convenient scripts to validate zero-shot performance of the pretrained model (see [Policy API Guide](getting_started/policy.md) and [RoboCasa Zero-Shot](examples/robocasa-gr1-tabletop-tasks/README.md)).\n4. Our repo provides examples of different configurations for training with different robot embodiments (see [`examples/`](examples/) and [Fine-tuning Guide](getting_started/finetune_new_embodiment.md)).\n5. Our repo provides convenient scripts for finetuning the pre-trained GR00T N1.6 model on user's data, and running inference, see [`examples`](examples).\n6. Our repo provides convenient scripts to run academic simulation benchmarks with finetuned checkpoints (see [LIBERO](examples/LIBERO/README.md), [SimplerEnv](examples/SimplerEnv/README.md), [RoboCasa](examples/robocasa/README.md)).\n7. The user will need to connect the `Gr00tPolicy` to the robot controller to execute actions on their target hardware.\n\n## What's New in GR00T N1.6\n\nGR00T N1.6 represents a significant upgrade over GR00T N1.5, with improvements in both model architecture and data leading to better performance in many aspects.\n\n### Model and Data Improvements\n\nArchitectural changes:\n- Base VLM: We use an internal NVIDIA Cosmos-Reason-2B VLM variant. The VLM supports flexible resolution and can encode images in their native aspect ratio without padding. The VLM is trained both general vision-language tasks and embodied reasoning tasks like next action prediction.\n- Uses 2x larger DiT (32 layers vs 16 layers in N1.5).\n- Removes N1.5's post-VLM 4-layer transformer adapter. Instead, unfreezes top 4 layers of the VLM during pretraining.\n- Predicts state-relative action chunks for most embodiments, rather than absolute joint angles or EEF positions.\n\nBeyond the N1.5 data mixture, the N1.6 pretraining data additionally includes several thousand hours of teleoperated data from:\n- Bimanual YAM arms\n- AGIBot Genie1\n- Simulated Galaxea R1 Pro on the BEHAVIOR suite\n- Whole-Body Locomanipulation with Unitree G1\n\nOther code-level improvements:\n- Faster dataloader with sharded dataloader support.\n- RTC and Async Policy Wrapper for inference (soon to release)\n- Simplified data processing pipeline with `processing_gr00t_n1d6.py`\n- Flexible Training configuration\n\n## Target Audience\n\nGR00T N1.6 is intended for researchers and professionals in robotics. This repository provides tools to:\n\n- Leverage a pre-trained foundation model for robot control\n- Fine-tune on small, custom datasets\n- Adapt the model to specific robotics tasks with minimal data\n- Deploy the model for inference\n\nThe focus is on enabling customization of robot behaviors through finetuning.\n\n## Installation Guide\n\n### Clone the Repository\n\nGR00T relies on submodules for certain dependencies. Include them when cloning:\n\n```sh\ngit clone --recurse-submodules https://github.com/NVIDIA/Isaac-GR00T\ncd Isaac-GR00T\n```\n\nIf you've already cloned without submodules, initialize them separately:\n\n```sh\ngit submodule update --init --recursive\n```\n\n### Set Up the Environment\n\nGR00T uses [uv](https://github.com/astral-sh/uv) for fast, reproducible dependency management.\n\n\u003e **Requirement:** uv **v0.8.4+** is needed to parse `[tool.uv.extra-build-dependencies]` in `pyproject.toml` (required for building `flash-attn`). For RTX-5090, this was tested with CUDA 12.8, `flash-attn==2.8.0.post2`, `pytorch-cu128`.\n\nAfter installing uv, create the environment and install GR00T:\n\n```sh\nuv sync --python 3.10\nuv pip install -e .\n```\n\n\u003e Note: CUDA 12.4 is recommended and officially tested. However, CUDA 11.8 has also been verified to work.\n\u003e In such cases, make sure to install a compatible version of `flash-attn` manually (e.g., `flash-attn==2.8.2` was confirmed working with CUDA 11.8).\n\nFor a containerized setup that avoids system-level dependency conflicts, see our [Docker Setup Guide](docker/README.md).\n\nFor training and inference hardware recommendations (RTX PRO Servers, DGX, Jetson AGX Thor), see the [Hardware Recommendation Guide](getting_started/hardware_recommendation.md).\n\n## Model Checkpoints\n\n### Base Models\nWe provide pre-trained base VLA model checkpoints. These checkpoints have been pre-trained on 10k+ hours of robot data and can be used for finetuning on downstream tasks.\n\n| Model | Use Case | Description | Checkpoint Path | Branch |\n| ----- | -------- | ----------- | --------------- | ------ |\n| GR00T N1.5 | Finetuning | Base [GR00T N1.5 model](https://research.nvidia.com/labs/gear/gr00t-n1_5/) (3B parameters) | [nvidia/GR00T-N1.5-3B](https://huggingface.co/nvidia/GR00T-N1.5-3B) | [n1.5-release](https://github.com/NVIDIA/Isaac-GR00T/tree/n1.5-release) |\n| GR00T N1.6 | Finetuning | Base [GR00T N1.6 model](https://research.nvidia.com/labs/gear/gr00t-n1_6/) (3B parameters) | [nvidia/GR00T-N1.6-3B](https://huggingface.co/nvidia/GR00T-N1.6-3B) | [main](https://github.com/NVIDIA/Isaac-GR00T) |\n\n### Finetuned Models\nWe also provide finetuned checkpoints for various robot platforms and benchmarks. These models are finetuned from the base models above and can be used directly for evaluation or as starting points for further finetuning.\n\n| Model | Base Model | Description | Checkpoint Path | Example |\n| ----- | ---------- | ----------- | --------------- | ------- |\n| GR00T-N1.6-bridge | [nvidia/GR00T-N1.6-3B](https://huggingface.co/nvidia/GR00T-N1.6-3B) | Fine-tuned on [Bridge dataset](https://rail-berkeley.github.io/bridgedata/) for WidowX robot on manipulation tasks | [nvidia/GR00T-N1.6-bridge](https://huggingface.co/nvidia/GR00T-N1.6-bridge) | [SimplerEnv](examples/SimplerEnv/README.md) |\n| GR00T-N1.6-fractal | [nvidia/GR00T-N1.6-3B](https://huggingface.co/nvidia/GR00T-N1.6-3B) | Fine-tuned on [Fractal dataset](https://www.tensorflow.org/datasets/catalog/fractal20220817_data) for Google robot on manipulation tasks | [nvidia/GR00T-N1.6-fractal](https://huggingface.co/nvidia/GR00T-N1.6-fractal) | [SimplerEnv](examples/SimplerEnv/README.md) |\n| GR00T-N1.6-BEHAVIOR1k | [nvidia/GR00T-N1.6-3B](https://huggingface.co/nvidia/GR00T-N1.6-3B) | Fine-tuned on [BEHAVIOR-1K](https://behavior.stanford.edu/) for Galaxea R1 Pro robot on loco-manipulation tasks | [nvidia/GR00T-N1.6-BEHAVIOR1k](https://huggingface.co/nvidia/GR00T-N1.6-BEHAVIOR1k) | [BEHAVIOR](examples/BEHAVIOR/README.md) |\n| GR00T-N1.6-G1-PnPAppleToPlate | [nvidia/GR00T-N1.6-3B](https://huggingface.co/nvidia/GR00T-N1.6-3B) | Fine-tuned for Unitree G1 loco-manipulation pick-and-place tasks | [nvidia/GR00T-N1.6-G1-PnPAppleToPlate](https://huggingface.co/nvidia/GR00T-N1.6-G1-PnPAppleToPlate) | [G1 LocoManipulation](examples/GR00T-WholeBodyControl/README.md) |\n| GR00T-N1.6-DROID | [nvidia/GR00T-N1.6-DROID](https://huggingface.co/nvidia/GR00T-N1.6-DROID) | Fine-tuned for DROID robot on manipulation tasks | [nvidia/GR00T-N1.6-DROID](https://huggingface.co/nvidia/GR00T-N1.6-DROID) | [DROID](examples/DROID/README.md) |\n\n\n\n## Quick Start\n\nWe can quickly start by downloading a pre-trained checkpoint and starting the policy server for any pretrained embodiement, e.g. GR1 embodiment.\n```bash\n# On GPU server: Start the policy server\nuv run python gr00t/eval/run_gr00t_server.py --embodiment-tag GR1 --model-path nvidia/GR00T-N1.6-3B\n```\n\nThen, refer to the [robocasa-gr1-tabletop-tasks](examples/robocasa-gr1-tabletop-tasks/README.md) for more details on how to rollout the policy with `GR1` embodiment.\n\n## Getting started with this repo\n\nWe provide accessible Jupyter notebooks and detailed documentation in the [`./getting_started`](getting_started) folder.\n\n## 1. Data Preparation\n\nPlease refer to the [data preparation guide](getting_started/data_preparation.md) for more details.\n\n## 2. Inference\n\nAfter data is prepared, the GR00T model can be used to generate output actions with the below simple inference script:\n\n```bash\nuv run python scripts/deployment/standalone_inference_script.py \\\n  --model-path nvidia/GR00T-N1.6-3B \\\n  --dataset-path demo_data/gr1.PickNPlace \\\n  --embodiment-tag GR1 \\\n  --traj-ids 0 1 2 \\\n  --inference-mode pytorch \\\n  --action-horizon 8\n```\n\nGR00T-N1.6-3B inference timing (4 denoising steps, single view):\n\n| Device | Mode | Data Processing | Backbone | Action Head | E2E | Frequency |\n|--------|------|-----------------|----------|-------------|-----|-----------|\n| RTX 5090 | torch.compile | 2 ms | 18 ms | 16 ms | 37 ms | 27.3 Hz |\n| H100 | torch.compile | 4 ms | 23 ms | 11 ms | 38 ms | 26.3 Hz |\n| RTX 4090 | torch.compile | 2 ms | 25 ms | 17 ms | 44 ms | 22.8 Hz |\n| Thor | torch.compile | 5 ms | 39 ms | 61 ms | 105 ms | 9.5 Hz |\n\nFor more details, please check our full [inference guide](scripts/deployment/README.md) for more details including faster inference with `TensorRT`\n\n## 3. Finetuning\n\n### Fine-tune on Pre-registered Post-train Embodiment Tags\n\nGR00T provides several pre-registered embodiment tags with ready-to-use configurations:\n\n- `LIBERO_PANDA`\n- `OXE_GOOGLE`\n- `OXE_WIDOWX`\n- `UNITREE_G1`\n- `BEHAVIOR_R1_PRO`\n\n**Example:** To finetune Libero-Spatial on GR00T N1.6, follow the instructions in the [Libero finetuning guide](examples/LIBERO/README.md#finetune-libero-spatial-dataset). We also provide simulation environment setup for evaluation linked with post-train checkpoints and benchmark numbers.\n\n### Fine-tune on Custom Embodiments (\"NEW_EMBODIMENT\")\n\nTo finetune GR00T on your own robot data and configuration, follow the detailed tutorial available at [`getting_started/finetune_new_embodiment.md`](getting_started/finetune_new_embodiment.md).\n\n#### Prerequisites\n\nEnsure your input data follows the **GR00T-flavored LeRobot v2 format**, and specify your modality configuration at `modality_config_path`.\n\n#### Run Fine-tuning Script\n```bash\n# Set number of GPUs\nexport NUM_GPUS=1\n\nCUDA_VISIBLE_DEVICES=0 uv run python \\\n    gr00t/experiment/launch_finetune.py \\\n    --base-model-path nvidia/GR00T-N1.6-3B \\\n    --dataset-path \u003cDATASET_PATH\u003e \\\n    --embodiment-tag NEW_EMBODIMENT \\\n    --modality-config-path \u003cMODALITY_CONFIG_PATH\u003e \\\n    --num-gpus $NUM_GPUS \\\n    --output-dir \u003cOUTPUT_PATH\u003e \\\n    --save-total-limit 5 \\\n    --save-steps 2000 \\\n    --max-steps 2000 \\\n    --use-wandb \\\n    --global-batch-size 32 \\\n    --color-jitter-params brightness 0.3 contrast 0.4 saturation 0.5 hue 0.08 \\\n    --dataloader-num-workers 4\n```\n\n\u003e For more extensive finetuning configuration, use `gr00t/experiment/launch_train.py` instead to launch the training process.\n\n### Recommended Fine-tuning Configuration\n\nFor optimal results, maximize your batch size based on available hardware and train for a few thousand steps.\n\n#### Hardware Performance Considerations\n\n**Fine-tuning Performance**\n- We recommend using 1 H100 node or L40 node for optimal finetuning performance\n- Other hardware configurations (e.g., A6000) will also work but may require longer training time\n- Optimal batch size depends on your hardware and which model components are being tuned\n\n#### Training Variance\n\nUsers may observe some variance in post-training results across runs, even when using the same configuration, seed, and dropout settings. In our experiments, we have observed performance differences as large as 5-6% between runs. This variance may be attributed to non-deterministic operations in image augmentations or other stochastic components. When comparing results to reported benchmarks, please keep this inherent variance in mind.\n\n## 4. Evaluation\n\nWe recommend a two-stage evaluation approach: open-loop evaluation followed by simulation evaluation to comprehensively assess model quality.\n\n### 4.1 Open-Loop Evaluation\n\nOpen-loop evaluation provides an offline assessment by comparing the model's predicted actions against ground truth data from your dataset.\n\n#### Running the Evaluation\n\nExecute the evaluation script with your newly trained model:\n```bash\nuv run python gr00t/eval/open_loop_eval.py \\\n    --dataset-path \u003cDATASET_PATH\u003e \\\n    --embodiment-tag NEW_EMBODIMENT \\\n    --model-path \u003cCHECKPOINT_PATH\u003e \\\n    --traj-ids 0 \\\n    --action-horizon 16  # ensure this is within the delta_indices of action's modality config.\n```\n\n#### Interpreting Results\n\nThe evaluation generates a visualization saved at `/tmp/open_loop_eval/traj_{traj_id}.jpeg`, which includes:\n- Ground truth actions vs. predicted actions\n- Unnormalized mean squared error (MSE) metrics\n\nThese plots provide a quick indicator of the policy's accuracy on the training dataset distribution.\n\n### 4.2 Closed-Loop Evaluation\n\nAfter validating performance through open-loop evaluation, test your model in closed-loop environments.\n\n#### Understanding the Policy API\n\nAfter training your model, you'll use the `Gr00tPolicy` class to load and run inference. The policy expects observations in a specific format (nested dictionaries with video, state, and language modalities) and returns actions ready for execution.\n\n**Quick Start with Server-Client Architecture:**\n\n```bash\n# On GPU server: Start the policy server\nuv run python gr00t/eval/run_gr00t_server.py \\\n    --embodiment-tag NEW_EMBODIMENT \\\n    --model-path \u003cCHECKPOINT_PATH\u003e \\\n    --device cuda:0 \\\n    --host 0.0.0.0 \\\n    --port 5555\n```\n\n```python\nfrom gr00t.policy.server_client import PolicyClient\n\npolicy = PolicyClient(host=\"localhost\", port=5555) # Connect to the policy server\nenv = YourEnvironment() # Create an environment\nobs, info = env.reset() # Reset the environment\nif not policy.ping(): # Verify connection\n    raise RuntimeError(\"Cannot connect to policy server!\")\naction, info = policy.get_action(obs) # Run inference\nobs, reward, done, truncated, info = env.step(action) # Execute the action\n```\n\n**Debugging with ReplayPolicy:**\n\nWhen developing a new environment integration or debugging your inference loop, you can use `ReplayPolicy` to replay recorded actions from an existing dataset. This helps verify that your environment setup, observation formatting, and action execution work correctly—without needing a trained model.\n\n```bash\n# Start server with ReplayPolicy (replays actions from dataset)\nuv run python gr00t/eval/run_gr00t_server.py \\\n    --dataset-path \u003cDATASET_PATH\u003e \\\n    --embodiment-tag NEW_EMBODIMENT \\\n    --execution-horizon 8  # should match the executed action horizon in the environment\n```\n\nThe server will replay actions from the first episode of the dataset. Use `policy.reset(options={\"episode_index\": N})` on the client to switch to a different episode.\n\n**For detailed documentation on:**\n- How to adapt the policy to your own environment\n- Server-client architecture for remote inference\n- Observation and action formats\n- Querying modality configurations\n- Batched inference\n- Troubleshooting common errors\n\nSee the complete [Policy API Guide](getting_started/policy.md).\n\n#### Evaluation Examples\n\nWe support evaluation on available public benchmarks and our internal benchmarks. Our evaluation framework uses a server-client architecture that communicates via RESTful API. Both the policy server and simulation environment client use the same IP (usually localhost) and port to run simulation evaluation.\n\nFor the policy server, we reuse the project root's uv environment (same as finetuning) to run `run_gr00t_server`. For simulation environment clients, we provide individual setup scripts to configure uv environments, as they typically conflict with each other when using a single shared environment.\n\nYou can use [the verification script](scripts/eval/check_sim_eval_ready.py) to verify that all dependencies and environments for simulation evaluation are properly configured.\n\nPlease refer to each benchmark link below for more details.\n\n**Zero-shot Evaluation** (evaluate without finetuning):\n- **RoboCasa**: [Instructions](examples/robocasa/README.md)\n- **RoboCasa GR1 Tabletop Tasks**: [Instructions](examples/robocasa-gr1-tabletop-tasks/README.md)\n\n**Finetuned Evaluation** (test after task-specific finetuning):\n- **G1 LocoManipulation**: [Instructions](examples/GR00T-WholeBodyControl/README.md)\n- **LIBERO**: [Instructions](examples/LIBERO/README.md)\n- **SimplerEnv**: [Instructions](examples/SimplerEnv/README.md)\n- **BEHAVIOR**: [Instructions](examples/BEHAVIOR/README.md)\n- **PointNav**: [Instructions](examples/PointNav/README.md)\n- **SO-100**: [Instructions](examples/SO100/README.md)\n\n\n# Contributing\n\nFor more details, see [CONTRIBUTING.md](CONTRIBUTING.md)\n\n\n## License \n\n```\n# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION \u0026 AFFILIATES. All rights reserved.\n# SPDX-License-Identifier: Apache-2.0\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n```\n\n\n## Citation\n\u003c!-- TODO: Update --\u003e\n[Paper Site](https://research.nvidia.com/labs/lpr/publication/gr00tn1_2025/)\n```bibtex\n@inproceedings{gr00tn1_2025,\n  archivePrefix = {arxiv},\n  eprint     = {2503.14734},\n  title      = {{GR00T} {N1}: An Open Foundation Model for Generalist Humanoid Robots},\n  author     = {NVIDIA and Johan Bjorck and Fernando Castañeda, Nikita Cherniadev and Xingye Da and Runyu Ding and Linxi \"Jim\" Fan and Yu Fang and Dieter Fox and Fengyuan Hu and Spencer Huang and Joel Jang and Zhenyu Jiang and Jan Kautz and Kaushil Kundalia and Lawrence Lao and Zhiqi Li and Zongyu Lin and Kevin Lin and Guilin Liu and Edith Llontop and Loic Magne and Ajay Mandlekar and Avnish Narayan and Soroush Nasiriany and Scott Reed and You Liang Tan and Guanzhi Wang and Zu Wang and Jing Wang and Qi Wang and Jiannan Xiang and Yuqi Xie and Yinzhen Xu and Zhenjia Xu and Seonghyeon Ye and Zhiding Yu and Ao Zhang and Hao Zhang and Yizhou Zhao and Ruijie Zheng and Yuke Zhu},\n  month      = {March},\n  year       = {2025},\n  booktitle  = {ArXiv Preprint},\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FNVIDIA%2FIsaac-GR00T","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FNVIDIA%2FIsaac-GR00T","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FNVIDIA%2FIsaac-GR00T/lists"}