Isaac Sim 6.1 General Availability

[rthaker01]

Isaac Sim 6.1 General Availability

NVIDIA Isaac Sim 6.1 is now generally available.

Platform Updates:

• Extension Hot-Reloading in Containers: Extensions modified inside the Isaac Sim container are now hot-reloaded on save, enabling faster iteration cycles when developing custom functionality without leaving the container.

Assets, Authoring, and Newton Capabilities:

• Asset Support for Multiple Physics Backends: Provided assets have been updated to the recommended asset structure, allowing them to be used seamlessly with either the PhysX or Newton physics backend.

• Improved Newton Authoring and Asset Conversion Tooling: Authoring robot assets for the Newton physics backend has been improved, with new UI workflows for applying MuJoCo and Newton schemas and detailed guidance on robot setup and best practices. Additionally, new Asset Transformer rules allow legacy PhysX assets, as well as assets imported from MJCF or URDF, to be easily converted for use across multiple physics backends.

• Newton Hydroelastic and SDF Contact Support: Isaac Sim now supports a hydroelastic contact model for the Newton physics backend, which resolves collisions over a contact patch rather than at discrete points, enabling more stable contact forces for dexterous manipulation tasks. The resulting contact data can also be used downstream for computing tactile feedback.

• System Identification (Experimental): New System Identification tooling allows simulation parameters to be tuned based on collected real-world robot data, helping users close the sim-to-real gap. Examples include joint drive and friction properties, actuator gains and delays, rigid body mass and inertia, and joint limits, with support for both PhysX and Newton (MuJoCo Warp solver). Please note that this functionality is experimental and may be removed or replaced with other tooling in future releases.

• Gains Tuner Support for Multiple Physics Backends: The Gains Tuner now supports Newton's MJWarp solver, allowing it to be used with either the PhysX or Newton physics backend. It can also now be used to tune the gains of Newton actuators.

• SimReady Validator: The SimReady Validator has been integrated into Isaac Sim, allowing robots (manipulators and grippers) as well as props to be validated against the open SimReady Foundation specifications, which define the requirements assets must satisfy to work reliably across rendering, simulation, and robotics workflows.

ROS 2 / Bridges / Policy Testing:

• ROS 2 Control Support: Isaac Sim now supports the standard ros2_control stack by hosting a Controller Manager directly inside the simulation and presenting the simulated robot as a hardware interface. Developers can test the same controllers, YAML configurations, and tooling they use on real hardware against simulation, then move to a physical robot by simply swapping the hardware interface.

• Policy Inference Improvements: When running trained policies in Isaac Sim, observation and action terms are now automatically constructed from Isaac Lab configuration files rather than needing to be handcrafted, simplifying the transition from policy training to testing.

• AGILE WBC Software-in-the-Loop Testing: Isaac Sim's documentation now links to a guide showing how the AGILE Whole-Body Control policy, trained in Isaac Lab, can be tested in Isaac Sim with software-in-the-loop on a Unitree G1. The policy is exported with LEAPP and run through Isaac ROS Deploy, allowing the same controller stack to be validated in simulation before being deployed to the real robot.

• Unified TF publishing: Transforms from multiple TF publisher nodes are now aggregated into a single merged message per simulation timestamp, so ROS 2 consumers always receive one coherent transform snapshot per tick. This results in better performance and cleaner ROS graphs, with fewer publisher nodes and no repeated child frames.

• ROS 2 H.265 Compressed Image Publishing: Isaac Sim now supports ROS 2 HEVC (H.265) compressed RGB publishing with hardware-accelerated encoding, in addition to the existing H.264 support. The accompanying decoder node has also been improved to automatically discover compressed image topics and republish them as raw images, simplifying setup on the subscriber side.

• Python Script Support in ROS 2 Launch: The Isaac Sim ROS 2 launch workflow now supports running custom Python scripts inside a standard GUI session after startup. Developers can automate scene setup and configuration directly from their launch files without writing a full standalone application, making it easier to bring up complete simulation workflows from a single launch command.

• Automatic ROS 2 Library Configuration for Standalone Scripts: The python.sh launcher now automatically configures Isaac Sim’s internal ROS 2 libraries when no ROS environment is sourced, allowing standalone scripts to use the ROS 2 bridge without manually setting environment environments.

• ZeroMQ Bridge Example: A new tutorial demonstrates how to connect Isaac Sim to an external process using ZeroMQ. This serves as a lightweight, dependency-free alternative to the ROS 2 bridge for streaming simulation data to and from Isaac Sim.

Robot Motion and Behavior:

• State Machines for Robot and Character Behavior: New tutorials demonstrate how to structure multi-step, reactive manipulation tasks using state machines and behavior trees, including the py_trees library and the new Omniverse Behavior Tree system. Omniverse Behavior Trees are authored as USD prims, execute on the simulation timeline, and can be inspected in a visual editor; beyond robot control, the system can also serve as a general decision-making framework, such as coordinating animation behaviors based on simulation state.

• Motion Generation API Adoption: Isaac Sim's manipulation examples and mobile robot controllers have been updated to leverage the Motion Generation API, providing a consistent controller interface across Isaac Sim's tooling and examples. Controllers built on this common, composable foundation are easier to learn from, adapt, and reuse in custom workflows.

Sensors:

• Camera Authoring Tutorial: A new tutorial demonstrates how to author a reusable camera sensor asset in USD, including lens distortion, tick rate, and supported resolutions and outputs. The resulting asset can be added to robots or environments for downstream synthetic data generation and software-in-the-loop testing workflows.

• Sensor Point Cloud Coloring: Point clouds from RTX Lidar and Radar sensors can now be colored by metadata (e.g. intensity, radial velocity) in the viewport with improvements to Isaac Sim Debug Drawing Extension.

Synthetic Data Generation and Teleoperation:

• Reconstructed Asset SDG Example: A new synthetic data generation example demonstrates how to generate synthetic data with an asset created using NVIDIA's 3D Object Reconstruction framework, which turns stereo video of a real object into a textured USD asset.

• TeleOperation Improvements: Teleoperation workflows have been expanded with several new capabilities. A new Visual Cues panel renders drop lines from each controller to a reference plane, improving depth perception when teleoperating from a 2D monitor. Added support for Isaac Teleop retargeting interface, allowing the trigger and squeeze inputs to independently drive the fingers of multi-finger hands. Locomotion additionally supports a new velocity-based drive mode, where the robot base is driven through physics for realistic contacts and collisions, and new built-in profiles demonstrate both the retargeting and velocity locomotion workflows.

Get started:

• Download Isaac Sim 6.1.

• Connect with the Isaac Sim community and share feedback through the NVIDIA Developer Forums and GitHub.