Update README.md

README.md

@@ -1,35 +1,62 @@
----
-title: VinRobotics
-emoji: 🤖
-colorFrom: blue
-colorTo: yellow
-sdk: static
-pinned: false
-license: apache-2.0
+# VinRobotics, Foundation AI
+
+**We make robotics foundation models fit.**
+
+Our work grounds robot learning in the parallel-computing paradigm: one policy, expressed so it
+executes on a server GPU, a Jetson-class board, an NPU, or a bare CPU, at the control rate the
+task demands and inside the power budget the robot carries. Everything published here is one
+half of that, either the runtime that executes a policy or the model that runs on it.
+
+Built and deployed on VR humanoids at [VinRobotics](https://vinrobotics.net) (Vingroup, Vietnam).
+
 ---
 
-# VinRobotics - Edge AI & Model Optimization
+## Runtime: [`vla.cpp`](https://github.com/VinRobotics/vla.cpp)
 
-We optimize and deploy **LLMs, ASR, VLM and VLA (Vision-Language-Action) models** on real-world systems.
+A C++/ggml inference engine for Vision-Language-Action models, built on
+[`llama.cpp`](https://github.com/ggml-org/llama.cpp). SmolVLA, π0 / π0.5, BitVLA, Evo-1 and
+GR00T N1.5/1.6/1.7 run under one runtime, each packaged as a single self-contained GGUF, with no
+Python and no PyTorch at inference time. Backends: CPU (AVX/NEON), Apple Silicon, CUDA from
+consumer GPUs down to Jetson, and Intel GPU via SYCL.
 
-## Featured Projects
+[arXiv:2606.08094](http://arxiv.org/abs/2606.08094) ·
+[Learn vla.cpp](https://fai-modelopt-tech.github.io/learn-vla-cpp/) ·
+[model bundles](https://huggingface.co/collections/vrfai/vlacpp-model-bundles-6a1e7e487bb2d85de81601a8)
 
-* **[vla.cpp](https://github.com/VinRobotics/vla.cpp)**
-  Native C++ inference runtime for Vision-Language-Action models, built for low-latency robotic deployment.
+## Compression: quantized checkpoints, and the recipes behind them
 
-* **[Model Quantization Recipes](https://github.com/VinRobotics/model-quantization-recipes)**
-  Practical recipes for quantizing and deploying LLM, ASR, VLM, and VLA models on real-world systems.
+Just landed: **Qwen3.8-27B** quantized three ways, ready to serve,
+[NVFP4](https://huggingface.co/vrfai/Qwen3.8-27B-NVFP4),
+[FP8](https://huggingface.co/vrfai/Qwen3.8-27B-FP8) and
+[FP8-dynamic](https://huggingface.co/vrfai/Qwen3.8-27B-FP8-dynamic).
 
+Deployment-validated NVFP4, FP8 and FP8-dynamic checkpoints for VLM, ASR and world models:
+[Qwen3.6](https://huggingface.co/collections/vrfai/qwen36-optimized-69e5d535c4b1465fd3afd611),
+[Qwen3-ASR](https://huggingface.co/collections/vrfai/qwen3-asr-optimized-69d8c8bd4cbad9c07b2cfbdb),
+[Gemma 4](https://huggingface.co/collections/vrfai/gemma-4-optimized-69e059ecb1c000d133d04694),
+[Cosmos](https://huggingface.co/collections/vrfai/cosmos-optimized-69f95c4177f946f2575fd48f).
+The method is open too:
+[model-quantization-recipes](https://github.com/VinRobotics/model-quantization-recipes) covers
+preparation through deployment-oriented validation, so a number can be reproduced rather than
+believed.
 
-## What we do
-- Optimization: quantization (INT8/INT4/FP8/NVFP4), pruning, distillation, ...
-- Deployment: VLLM, TensorRT, ONNX Runtime, edge runtimes  
-- Systems: real-time pipelines (vision, audio, language, action)
+---
+
+## Maintainers and contributors
+
+- Khanh Nguyen, [GitHub](https://github.com/khanhnd61-vr) · [HF](https://huggingface.co/khanhnd61)
+- Hung Ho, [GitHub](https://github.com/hungho77) · [HF](https://huggingface.co/hungho77)
+- Duc Minh Nguyen, [GitHub](https://github.com/ducido) · [HF](https://huggingface.co/ducido)
+- Quang Nguyen, [HF](https://huggingface.co/quangnd58)
+- Anh Huynh, [HF](https://huggingface.co/hqanh308)
+- Pham Tri Quang, [HF](https://huggingface.co/twanghcmut)
+- An Thai Le, Director of Foundation AI, [GitHub](https://github.com/anindex) · [HF](https://huggingface.co/anindex)
+
+---
 
-## Focus
-- Edge devices (Jetson, SoCs)  
-- Robotics & VLA systems  
-- Latency, stability, deployability  
+[vinrobotics.net](https://vinrobotics.net) ·
+[github.com/VinRobotics](https://github.com/VinRobotics) ·
+[LinkedIn](https://www.linkedin.com/in/vinrobotics) ·
+[support@vinrobotics.net](mailto:support@vinrobotics.net)
 
-## Philosophy
-Optimization = **model + runtime + system**
+Issues and PRs on the GitHub repos. Model-specific questions in the discussion tab of each repo here.