invisible fΦrce Knowledge hub
Tools

Which robotic simulators
exist today?

Three simulators dominate robotics today: NVIDIA Isaac Sim for large-scale training and photorealistic rendering, MuJoCo for accurate physics and research, and Gazebo for testing software built on ROS. A handful of others matter for specific jobs.

NVIDIA Isaac Sim

What it is. NVIDIA's simulator built on its Omniverse platform. It draws photorealistic images using RTX ray tracing, simulates physics with PhysX (and, experimentally, the new Newton engine), and connects to ROS 2. Its companion, Isaac Lab, is the framework for training thousands of robots in parallel on a GPU.[1][2][9]

Best for. Reinforcement learning at scale, generating labelled synthetic images, and building digital twins of factories. Invisible Force's modules plug into Isaac Sim and Newton.

Physics engines. Isaac Sim is a simulator, not a physics engine: it runs PhysX by default and, since version 6.x, the new Newton engine as an option. Both are described in What is a physics engine?[2][5]

Status. Isaac Sim 6.1.0 was released on 10 September 2026; Isaac Lab 3.0 is in beta. The source is open under the Apache 2.0 licence, but it needs an NVIDIA GPU.[1][2]

MuJoCo

What it is. "Multi-Joint dynamics with Contact", a physics engine and simulator now maintained by Google DeepMind and free under Apache 2.0. It is known for careful, stable modelling of contact, and for being fast on an ordinary CPU. Two GPU versions exist: MuJoCo Warp, developed with NVIDIA, and MJX, which also runs on other hardware.[3][4][10]

Best for. Control research, legged robots, and anywhere physics accuracy matters more than pretty pictures. Invisible Force's battery demo runs in MuJoCo.

Status. MuJoCo 3.13.0 and MuJoCo Warp 3.13.0 were released on 9 September 2026. MuJoCo Warp is also the main solver inside the Newton engine.[3][4][5]

Gazebo

What it is. The long-standing open-source simulator from Open Robotics, tightly integrated with ROS 2, the software framework most robots are built on. It focuses on sensors, mobile robots and testing the whole software stack rather than on raw physics speed; it can use several physics engines underneath.[6]

Best for. Checking that a robot's ROS software works before putting it on hardware. It does not run many robots in parallel on a GPU, so it is rarely used for reinforcement learning.[6]

Status. The current long-term release is Gazebo Jetty, September 2025, supported until 2031.[6]

Other simulators, in one line each

For a fuller comparison, two 2025 surveys compare simulators feature by feature.[14][15]

References

  1. NVIDIA, GitHub. Isaac Sim releases. September 2026.
  2. NVIDIA. Isaac Sim 6.1.0 release notes. September 2026.
  3. Google DeepMind, GitHub. MuJoCo 3.13.0 release. September 2026.
  4. Google DeepMind and NVIDIA. MuJoCo Warp (repository). September 2026.
  5. Newton project, Linux Foundation. Newton releases. September 2026.
  6. Open Robotics, Discourse announcement. Gazebo Jetty released. September 2025.
  7. Genesis AI, GitHub. Genesis world releases. September 2026.
  8. Toyota Research Institute. Drake v1.57.0 release notes. September 2026.
  9. NVIDIA, GitHub. Isaac Lab releases. July 2026.
  10. Google DeepMind. MuJoCo documentation: MJX. 2026.
  11. Hao Su Lab, UC San Diego, GitHub. ManiSkill releases. April 2026.
  12. bulletphysics, GitHub. Bullet Physics releases. April 2022.
  13. Cyberbotics, GitHub. Webots R2025a release. February 2025.
  14. Wong, Kang, Bai and Zhang, arXiv 2505.01458. A Survey of Robotic Navigation and Manipulation with Physics Simulators in the Era of Embodied AI. May 2025, revised June 2026.
  15. Long, Zhao, Zhang et al., arXiv 2507.00917. A Survey: Learning Embodied Intelligence from Physical Simulators and World Models. July 2025, revised September 2025.

Every link was opened and checked in September 2026. Links open in a new tab.