keivalya

keivalya / nemotron-vla

Public

Open source VLA Model powered by NVIDIA Foundation Models

25
4
100% credibility
Found Feb 17, 2026 at 20 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Jupyter Notebook
AI Summary

This repository provides tools to collect expert demonstration data from simulated robot manipulation tasks and train a vision-language-action AI model to perform them based on natural language instructions.

How It Works

1
🔍 Discover Robot Teacher

You find this fun project that helps teach AI robots to do simple tasks like picking up objects or opening doors using words and sights.

2
📹 Gather Example Clips

You collect short video clips of perfect robot performances for many chores, each paired with easy instructions like 'push the ball into the basket'.

3
🧠 Boost with Smart Insights

The clips get enhanced with clever vision understanding from pictures and word meanings from instructions to prepare rich learning material.

4
🚀 Train the Robot Brain

You start training where the AI blends what it sees, hears in words, and feels to learn smooth actions just like the examples.

5
🎮 Give Commands and Watch

You tell the robot a task in plain words, like 'close the drawer', and see it act in the simulated world.

✅ Robot Nails the Tasks

Your AI robot confidently handles instructions, succeeding at chores and opening doors, feeling like magic for everyday robot smarts.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 20 to 25 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is nemotron-vla?

Nemotron-VLA is an open source VLA model for robotics that lets you train vision-language-action policies on MetaWorld manipulation tasks using NVIDIA foundation models. It collects multitask expert demonstrations with natural language instructions like "push the object to the goal," extracts features via RADIO vision and Nemotron language encoders, then trains a diffusion-based action head—all in Jupyter notebooks. Developers get a full pipeline from data gen to inference videos, tackling sim robotics control with hybrid VLA github setups.

Why is it gaining traction?

It stands out by freezing massive NVIDIA foundation models (RADIO for 3D VLA github vision, Nemotron Nano for lang) while training only a lightweight fusion and diffusion policy, slashing compute needs versus from-scratch VLA model training. The end-to-end Jupyter notebook flow—data collection, feature extraction, training, eval—makes it an awesome VLA github starter for cot vla github or vla rl github experiments, with easy video outputs for quick iteration. Low overhead beats heavy VLA model Huggingface ports or custom RL baselines.

Who should use this?

Robotics devs prototyping VLA model robot policies in simulation, like MetaWorld task generalization. Researchers tuning VLA adapter github for lang-conditioned manipulation or hybrid VLA github with foundation models. Teams needing a VLA model tutorial for vla model size benchmarks or open source VLA OS github baselines.

Verdict

Solid prototype for VLA model ai in robotics, but 1.0% credibility score and 18 stars signal early-stage maturity—expect tweaks for production. Grab it if you're spinning up VLA model open source experiments fast; skip for battle-tested alternatives.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.