NVIDIA Launches Isaac Robotics Platform 2.0 with Revolutionary Sim-to-Real Transfer
Source: NVIDIA Developer Blog
NVIDIA has unveiled Isaac Platform 2.0 at GTC 2026, a comprehensive update to its robotics development platform that promises to make building and deploying intelligent robots dramatically faster and easier.
Key Features
Isaac Sim 2.0
The updated simulation environment features:
- Photorealistic rendering using RTX path tracing for ultra-realistic visual training data
- Improved physics with GPU-accelerated deformable body simulation
- Digital twin sync — real-time mirroring between physical robots and their simulated counterparts
- 10x performance — simulate 100,000+ robots simultaneously on a single DGX system
Isaac Lab 2.0
The reinforcement learning framework gets major upgrades:
- One-click sim-to-real — automated domain randomization and reality gap analysis
- Pre-built task libraries — 50+ ready-to-use manipulation and locomotion tasks
- Multi-GPU training — linear scaling across up to 8 GPUs
- Foundation policy — a pre-trained manipulation policy that can be fine-tuned in hours instead of days
Isaac Manipulator Foundation Model
The headline announcement: a new foundation model for robot manipulation trained on 10 million simulated grasps:
- Picks up novel objects with 94% success rate (zero-shot)
- Handles transparent, reflective, and deformable objects
- Runs at 30 Hz on NVIDIA Jetson Orin
- Open weights available for research use
Pricing
| Tier | Price | Includes |
|---|---|---|
| Developer | Free | Isaac Sim, Isaac Lab, community support |
| Professional | $2,000/year | Priority support, advanced features |
| Enterprise | Custom | Dedicated support, custom model training |
Partner Ecosystem
Over 200 robotics companies have committed to building on Isaac Platform 2.0, including:
- Universal Robots — cobot programming and simulation
- Agility Robotics — Digit humanoid training
- Clearpath Robotics — autonomous mobile robot development
- Franka Emika — manipulation research
Availability
Isaac Platform 2.0 is available immediately for download from the NVIDIA Developer portal, with the foundation model weights releasing in March 2026 under an open research license.
This launch positions NVIDIA as the dominant platform provider for robotics AI development, much as it has become for AI training infrastructure.
Related Posts
Sim-to-Real Transfer: Bridging the Gap Between Virtual Training and Physical Robots
The complete guide to sim-to-real transfer for robotics — how to train robot policies in simulation and deploy them on real hardware. Covers domain randomization, system identification, and the latest gap-closing techniques.
Reinforcement Learning for Robot Control: A Deep Dive into Training Robots Through Trial and Error
How reinforcement learning enables robots to learn complex physical tasks — from walking and grasping to playing table tennis. Covers sim-to-real transfer, reward shaping, and the latest RL algorithms for robotics.