Today, NVIDIA announced the availability of the 2022.1 release of NVIDIA Isaac Sim. As a robotics simulation and synthetic data generation (SDG) tool, this NVIDIA Omniverse application accelerates the development, testing, and training of AI in robotics.
With Isaac Sim, developers can generate production quality datasets to train AI-perception models. Developers will also be able to simulate robotic navigation and manipulation, as well as build a test environment to validate robotics applications continually.
The latest version advances the age of AI robots with new tools like NVIDIA Isaac Cortex, a decision framework for training collaborative robots (cobots), and Isaac Gym, a GPU-accelerated reinforcement learning (RL) framework. NVIDIA Isaac Replicator, a set of synthetic data generation tools, APIs, and workflows, has also been updated with new capabilities to generate industrial environments for SDG procedurally.
NVIDIA Isaac Sim 2022.1 release highlights
Isaac Cortex: Program cobot tasks as easily as programming game AI. Leverage this decision framework for cobots to develop task-aware and adaptive skills. Using its belief representation of the world, analogous to the robot’s brain, real or simulated data can be used as inputs and the resulting actuations will be generated.
Isaac Gym: Train robots in minutes instead of weeks. Train complex robotic skills using RL. The Isaac Gym is a GPU-accelerated tool that keeps the entire RL training workflow on the GPU, which is critical to reduce training time.
Omnigraph: Simplify application development and debugging with visual programming. Build robotic applications by visually connecting compute nodes together in this Omniverse visual programming and scripting environment. Robotic applications tend to be very modular and lend themselves well to visual programming.
Isaac Sim/Gazebo Connector: Move between both simulators depending on tasks. ROS developers using Gazebo can import simulation assets into Isaac Sim for tasks like generating synthetic datasets or high-fidelity rendering. Additionally, multiple gazebo simulations can stay live synched by connecting to Omniverse’s nucleus server.
- Windows Support (limited)
- New Robots
- Quadrupeds: A1, GO1, Anymal
- AMR: Obelix
- New modular warehouse and conveyor assets
- New ROS pipelines implemented in Omnigraph
Training AI with synthetic data
Isaac Replicator is the synthetic data generation tool in Isaac Sim. Synthetic data is very useful in robotics to bootstrap training, address long-tail dataset challenges, and provide unavailable real-world data like speed and direction from synthetic video. Autonomous machines require synthetic data in training to ensure model robustness.
In the latest release, a new SDG feature called SceneBlox was added to generate scenes procedurally. SceneBlox can be used to create industrial environments like warehouses automatically. New examples were also added that demonstrate how to generate synthetic data and train a pose estimation model using Replicator.