How GPT-6 Astra and NVIDIA Isaac Tools Speed Up Robot Skill Training from Simulation to the Real World

Building robots is difficult, and creating simulation worlds for training and testing requires time and specialized expertise. This presentation shows how GPT-6 Astra lets developers describe assets, environments, and tasks with a simple prompt, then builds the training world and prepares it for robot skill training using NVIDIA tools. It also covers the path from simulation to real-world deployment.

■ Simulation World Creation
- Describing assets, environments, and tasks through prompts and data
- Building the training world with NVIDIA open libraries and models
- Replaying real-world motion and tuning physics for training readiness
■ Skill Training and Evaluation
- Training and refining robot skills with Isaac Lab and Newton
- Evaluating skills at scale with Isaac Lab Arena
- Running all tools locally on NVIDIA DGX Station for faster iteration
■ Simulation-to-Real Deployment
- Skills applied across homes, hospitals, stores, factories, and more
- Agile 1 SN, a full-stack reference robot built on NVIDIA Isaac Groot with Jetson Thor
- A starting point for developers to train and deploy new skills

ロボット開発やシミュレーション-basedのスキル習得に関心がある開発者、NVIDIAのエコシステムを確認したい方を対象としています。プロンプトから訓練環境を構築し、スキルの学習・評価・実機展開に至る一連の流れを把握でき、自らの開発環境を整える際の参考になります。