Unity launches Simulation Pro early access for robotics

Unity opened early access to Simulation Pro, a supported robotics package for Unity Industry customers that imports a robot description, simulates sensors, connects to ROS 2, and can run many copies on Linux without a screen.

SOFTWARE desk — a supported simulation package with a robot-description importer and ROS 2 is the plumbing physical AI teams were rebuilding on every project. Physical AI means software that has to run on a machine in the real world, not only answer a question on a screen.

What a team needs in order to use it. The post says Simulation Pro is compatible with Unity 6.3 and newer and requires a Unity Industry license. Unity 6.3 is the editor release the package runs on. Newer means that release or a later one. Unity Industry is the paid product line for companies. Early access means those customers can use the package before a general release. The post says it is available beginning Tuesday, September 22. The registration link on the page is a Unity contact form. The page does not print a price. This desk did not submit the form.

What the package includes, in the post’s words. It is a fully supported robotics toolkit for importing, simulating, and iterating on robotics systems in Unity. Four pieces. A URDF importer: URDF means Unified Robot Description Format, a standard file that describes a robot’s parts and how the joints connect. The importer is drag-and-drop, and the kinematic chain stays intact. A kinematic chain is that linked set of joints, so the arm still bends the way the file says. Sensor simulation out of the box, covering LiDAR, image, and IMU. LiDAR is a sensor that measures distance with pulses of light. An image sensor, here, is a camera. An IMU is an inertial measurement unit, a sensor that tracks motion and tilt. ROS 2 compatibility, connecting Unity to systems that run Robot Operating System 2, the current generation of that open toolkit. A headless Linux build target, so teams can run simulations without a graphical interface and scale them in parallel on hardware they already have. Headless means no screen. Parallel means many copies at once. File that list as the post’s. This desk did not run a sensor.

Why the company says it matters, in the chief operating officer’s words. Alex Blum said robotics teams came to Unity because they needed real-time rendering, accurate physics, and sensor simulation they could trust. Real-time rendering means drawing the scene fast enough to keep up with those sensors. He said they wanted a foundation to move robots from the lab to the real world, that Simulation Pro is that foundation, and that Unity Industry customers can access it today. File the name, the title, and those sentences as the post’s. A quote is not a count of customers.

The driving example the post cites, and what it is not. TIER IV built AWSIM, an open-source simulator developed with Unity, to test Autoware, open-source software for a car that drives itself. The post says some AWSIM scenarios call for eight or more cameras running at once, while keeping the sensor picture faithful, and that ordinary tricks for making a game run faster were not enough. The team built its own optimizations in Unity: staggered rendering across cameras, so the frames do not all spike at once; time-scale controls, to spend more computing on each frame; lighter assets; and the option to use different rendering pipelines in one project, trading accuracy against frame rate. A rendering pipeline is the path from the 3D scene to the picture a camera produces. The post says the team also used Unity’s job system and built-in physics to hold real-time performance across the vehicle, the traffic, and the pedestrians. For hardware-in-the-loop testing, the team analyzed raw sensor packets and reproduced equivalent output from Unity, so a camera setup could be tested before anything went on a physical test vehicle. Hardware-in-the-loop means the real device sits in the test, and the simulated sensors feed it. The post says that cut test-vehicle hours and on-road cost. These lines are Unity’s account of TIER IV’s earlier work in Unity. The post does not say TIER IV installed Simulation Pro on 22 Sep. Eight or more cameras is the post’s scenario line. This desk did not count them.

The factory example, still the post. The Korea Institute of Industrial Technology, shortened to KITECH, needed training data for how people and robots share a manufacturing floor. The post says a well-run plant is laid out so that close moments between a worker and a machine do not happen, and that staging those moments in order to film them is not a reasonable option. To measure proximity, video, depth, robot joints, and human pose had to exist at the same moment, so distance and safety-zone labels could be calculated the same way each time. KITECH built a digital twin of a metal casting plant from point-cloud scans, because no CAD source existed. A digital twin is a software copy of the plant. A point cloud is a 3D scan made of many measured points. CAD is the computer drawing of the factory, which the post says was missing. The team then built physics, sensors, and simulation on that twin. The sensor rig used custom sensor components matching the real rig, so every observation shared one simulation clock. They used Unity’s Articulation Body components, replaying recorded joint paths so inertia and contact were computed together, and the motion was physically valid rather than merely plausible. The post says that fidelity let KITECH automatically generate over 100,000 pre-labeled training frames in 14 weeks. A frame, here, is one labeled picture from the simulation. Pre-labeled means the distance and safety-zone marks were computed, not drawn by hand. Over 100,000 and 14 weeks are the post’s figures. This desk did not audit the frame count. The post does not say KITECH is an early-access customer of the new package.

The surgery example, still the post. Medtronic built a data-logging and playback pipeline in Unity around a digital twin of its Hugo robotic-assisted surgery system, capturing 195 procedures across five global sites. 195 and five are the post’s figures. Matthew Euliano, senior software engineer at Medtronic, said Unity is their first stop for prototypes so they can put things in front of surgeons quickly. The post says the recordings became a reusable dataset. Instead of waiting for operating-room access, teams could load a recorded procedure, scrub through it, isolate the moment, and inspect it from any angle. The same data supported deterministic replays — the same recording plays back the same way each time — to train new AI models and to inform the design of later devices. File that as Unity’s account of Medtronic’s work. This desk did not watch a procedure. The post does not say Hugo now runs on Simulation Pro.

Where the post says to start. Early access begins Tuesday, September 22. The registration link is a contact form. The line the post labels as package documentation goes to a Unity Discussions thread. A robotics learning pathway is on Unity Academy. The post says Unity is especially interested in teams running robot validation, sensor testing, or synthetic data generation at scale, across manufacturing, logistics, warehousing, and other advanced robotics. Synthetic data means pictures and sensor readings made in simulation and used to train a model. File that interest line as Unity’s. This desk did not open the Academy course.

Leave the blanks the page left open. There is no price. There is no customer count for the new package. There is no hour on the post. There is no benchmark, and this filing does not invent one. The three case studies are examples of earlier Unity robotics work the post cites. They are not a claim that each team switched to Simulation Pro on this announcement. A public blog post is not a test this desk ran.

Plain English for the rest of the card: Simulation Pro = the supported robotics package in this early access. Unity Industry = the paid product line the license has to cover. Unity 6.3 = the editor release the package runs on. early access = open to those customers before a general release. URDF = Unified Robot Description Format, the file that describes the robot. kinematic chain = the linked joints, kept intact by the importer. LiDAR = a sensor that measures distance with light. IMU = inertial measurement unit, a sensor for motion and tilt. ROS = Robot Operating System, the open toolkit for robot software. ROS 2 = the current generation of that toolkit. headless = a build that runs without a screen. parallel = many copies at once. digital twin = a software copy of a plant, a car’s surroundings, or a surgical robot. point cloud = a 3D scan made of measured points. CAD = the computer drawing of a factory. hardware-in-the-loop = the real device sits in the test, fed by simulated sensors. synthetic data = training pictures and sensor readings made in simulation. This filing is the 22 Sep early-access announcement. It is not a price and not a benchmark.

PRIMARY here: Unity’s 22 Sep 2026 blog post, “From simulation to real-world deployment: Unity Simulation Pro early access,” by Isaac Seah — Tier A PRIMARY, the company’s own record. The early-access launch, the Unity 6.3 and newer line, the Unity Industry license, the URDF importer with the kinematic chain intact, the LiDAR, image, and IMU sensors, the ROS 2 line, the headless Linux build target, the Blum quote, the TIER IV AWSIM and Autoware example with eight or more cameras, the KITECH digital twin and the over-100,000 frames in 14 weeks, the Medtronic Hugo line of 195 procedures across five sites, the Euliano quote, the Tuesday, September 22 availability, the contact-form registration, the Discussions thread the post uses for package documentation, and the Unity Academy pathway are the post’s. NOT claimed: a price, a customer count for the new package, an hour stamp, a benchmark, that TIER IV, KITECH, or Medtronic installed Simulation Pro on 22 Sep, that this desk installed the package, opened the Academy course, or submitted the form, a stock tip, or investment advice. Distinct from the already-filed nvidia-isaac-ros-5, intrinsic-core-open-source, and cognex-realsense-500m.