Make your own Robot with Arduino Ventuno Q

I built a robot arm that picks up rubber ducks using Arduino's most powerful board and an Vision Language Action model that learned a task from just 50 demonstrations. In this video, I fine-tune HuggingFace SmolVLA on the pick and place dataset I created myself. Then I run the inference on Arduino Ventuno Q - a Qualcomm QCS8275 board with 40 TOPS NPU - connected to SO-101 robotic arm. Useful links: https://store.arduino.cc/products/ventuno-q https://youtu.be/49UT-FD4_Go https://huggingface.co/docs/lerobot/main/en/il_robots https://huggingface.co/AIWintermuteAI/smolvla-ducks-gguf https://github.com/AIWintermuteAI/vla.cpp https://medium.com/correll-lab/fine-tuning-smolvla-for-new-environments-code-included-af266c56d632 https://github.com/huggingface/lerobot/issues/2354 https://github.com/huggingface/lerobot/issues/2258 https://huggingface.co/HuggingFaceVLA/smolvla_libero Most of inference/lerobot-vla.cpp plugin code is in spaghetti state: I need some time and effort to make it publishable, initially CPU inference. Comment to cheer me up!