Spotlight
April 8, 2026
UE5 is becoming the platform of choice for robotics simulation
The simulation industry is undergoing a major pivot, moving from building tools that train humans to building tools that train machines.
This requires AI systems that operate, perceive, and interact directly with the real world. Enter Physical AI.
By combining AI models with sensors, cameras, and actuators, these systems, such as robots, autonomous vehicles, and drones, understand, reason, and act in real time, adapting to unpredictable environments.
Unreal Engine is becoming the go-to platform for robotics teams, not just as a simulator, but as a synthetic data factory and real-time autonomy platform powering the next generation of robotics systems.
Training a robot to operate in the real world requires vast amounts of data: annotated images for perception, physics interactions for control policies, and edge-case scenarios for safety validation.
Collecting this data in the physical world is slow, expensive, dangerous, and fundamentally limited: you can’t stage a thousand car crashes, simulate a hurricane inside a warehouse, or replay a surgical complication on demand.
This is why simulation has become the backbone of modern Physical AI development. With simulation non-negotiable, the real question is, “Which platform delivers the best perception performance at the lowest cost and fastest iteration speed?”
For robotics teams, Unreal Engine is increasingly the answer. That’s because in robotics simulation, photorealism is more than a nice-to-have: it is a technical requirement.
The engine’s powerful graphical capabilities make it the best option for closing the “sim-to-real gap”: the gap between what a model sees in simulation and what it encounters on a physical robot.
The logic is simple: the more your synthetic images look like real photographs, the less additional work your model needs to generalize from simulation to reality.
Unreal Engine 5 introduced three technologies that fundamentally changed the game on this front not only delivering photorealism, but doing so in real time.
Nanite virtualized micropolygon geometry renders extremely high resolution assets in real time ideal for CAD imports, photogrammetry scans, and dense industrial environments.
Lumen dynamic global illumination calculates virtually infinite light bounces and indirect reflections in real time, enabling physically accurate lighting that reacts to changes in direct lighting (such as a light turning on) or geometry (such as a door opening).
And hardware ray tracing, including NVIDIA’s RTX-accelerated branch, delivers physically accurate reflections, shadows, and material responses.
General Robotics uses Unreal Engine to create realistic virtual environments for testing and demonstrating robotic autonomy systems, bringing scalable simulation and visual validation to robotics platforms.
“Unreal Engine’s photorealism and world control enables the prediction of real-world behavior across varying scenarios—and the agentic autonomy makes it scalable,” says Sai Vemprala, Co-Founder and CTO of General Robotics.
This requires AI systems that operate, perceive, and interact directly with the real world. Enter Physical AI.
By combining AI models with sensors, cameras, and actuators, these systems, such as robots, autonomous vehicles, and drones, understand, reason, and act in real time, adapting to unpredictable environments.
Unreal Engine is becoming the go-to platform for robotics teams, not just as a simulator, but as a synthetic data factory and real-time autonomy platform powering the next generation of robotics systems.
Photorealism as a technical requirement
Training a robot to operate in the real world requires vast amounts of data: annotated images for perception, physics interactions for control policies, and edge-case scenarios for safety validation.Collecting this data in the physical world is slow, expensive, dangerous, and fundamentally limited: you can’t stage a thousand car crashes, simulate a hurricane inside a warehouse, or replay a surgical complication on demand.
This is why simulation has become the backbone of modern Physical AI development. With simulation non-negotiable, the real question is, “Which platform delivers the best perception performance at the lowest cost and fastest iteration speed?”
For robotics teams, Unreal Engine is increasingly the answer. That’s because in robotics simulation, photorealism is more than a nice-to-have: it is a technical requirement.
The engine’s powerful graphical capabilities make it the best option for closing the “sim-to-real gap”: the gap between what a model sees in simulation and what it encounters on a physical robot.
The logic is simple: the more your synthetic images look like real photographs, the less additional work your model needs to generalize from simulation to reality.
Unreal Engine 5 introduced three technologies that fundamentally changed the game on this front not only delivering photorealism, but doing so in real time.
Nanite virtualized micropolygon geometry renders extremely high resolution assets in real time ideal for CAD imports, photogrammetry scans, and dense industrial environments.
Lumen dynamic global illumination calculates virtually infinite light bounces and indirect reflections in real time, enabling physically accurate lighting that reacts to changes in direct lighting (such as a light turning on) or geometry (such as a door opening).
And hardware ray tracing, including NVIDIA’s RTX-accelerated branch, delivers physically accurate reflections, shadows, and material responses.
General Robotics uses Unreal Engine to create realistic virtual environments for testing and demonstrating robotic autonomy systems, bringing scalable simulation and visual validation to robotics platforms.
“Unreal Engine’s photorealism and world control enables the prediction of real-world behavior across varying scenarios—and the agentic autonomy makes it scalable,” says Sai Vemprala, Co-Founder and CTO of General Robotics.
Courtesy of General Robotics
The engine is the foundation for agentic robot evaluation: an agent orchestrates and executes evaluations of robot AI behaviors across different environments and conditions, finding edge cases and assessing performance before deployment.
Not only that, but because Unreal Engine runs on macOS, robotics engineers can run their simulators natively on their laptops rather than logging into Linux via SSH. This brings the advantages of lower latency, better graphics performance, and a simplified workflow.
In many manipulation and locomotion tasks, inaccuracies in contact dynamics or actuator behavior cause failure long before visual differences do.
In this context, Unreal Engine’s Chaos Physics system with double precision at its core is powerful for scalable training, perception simulation, and real-time interaction within photorealistic environments.
Engineers also have the option to model the accurate physical behavior of systems using third-party tools such as AGX Dynamics for Unreal Engine, a high-fidelity real-time physics simulation system developed by Algoryx.
Not only that, but because Unreal Engine runs on macOS, robotics engineers can run their simulators natively on their laptops rather than logging into Linux via SSH. This brings the advantages of lower latency, better graphics performance, and a simplified workflow.
Physics simulation
Photorealism alone does not close the sim-to-real gap. Physics fidelity (accurate contact, friction, deformables, actuator dynamics), sensor realism (rolling shutter, LiDAR intensity, IMU drift, latency), and proper control-loop modeling are equally critical.In many manipulation and locomotion tasks, inaccuracies in contact dynamics or actuator behavior cause failure long before visual differences do.
In this context, Unreal Engine’s Chaos Physics system with double precision at its core is powerful for scalable training, perception simulation, and real-time interaction within photorealistic environments.
Engineers also have the option to model the accurate physical behavior of systems using third-party tools such as AGX Dynamics for Unreal Engine, a high-fidelity real-time physics simulation system developed by Algoryx.
Courtesy of Algoryx Simulation
AGX enables physically accurate simulation of complete machines, including multibody mechanics, hydraulic systems, drivetrains, engines, tracks or wheels, and dynamically deformable terrain.
“Autonomous machines must reason about physics,” explains Kenneth Bodin, CEO and Co-Founder of Algoryx. “If the simulation does not capture how the machine actually interacts with the world, the AI will learn the wrong behaviors. By combining engineering-grade physics simulation from AGX with Unreal Engine’s photorealistic environments, we can build digital worlds where autonomous machines can be trained, tested, and validated at scale.”
For engineers, this pipeline is scalable and automated. RealityScan exports critical registration and pose data while supporting standard computer vision (CV) formats (such as Colmap). Its state of the art (SOTA) image alignment provides the backbone for creating Gaussian splats—high-fidelity, photorealistic assets that can be produced with relatively low effort.
RealityScan can run on Linux servers and be controlled by the Remote Command Plugin or CLI scripts, producing automated photogrammetry pipelines that process scans continuously across multiple machines and send the results directly into Unreal Engine.
So instead of requiring humans to label images after they’re captured, the engine can automatically output highly accurate labels at the same time it renders the image.
“Autonomous machines must reason about physics,” explains Kenneth Bodin, CEO and Co-Founder of Algoryx. “If the simulation does not capture how the machine actually interacts with the world, the AI will learn the wrong behaviors. By combining engineering-grade physics simulation from AGX with Unreal Engine’s photorealistic environments, we can build digital worlds where autonomous machines can be trained, tested, and validated at scale.”
Merge and reconstruct different data sources with RealityScan
By combining cutting-edge photogrammetry and real-time technology, robotics teams can build a powerful pipeline to build realistic virtual environments. A case in point is the well-established workflow between RealityScan and Unreal Engine, which rapidly converts real-world sites and objects into digital twins for robotics simulation. RealityScan now supports SLAM point clouds as well as traditional photogrammetry and LiDAR, eliminating the environment creation bottleneck and enabling the rapid generation of synthetic training and regression environments that closely match deployment domains. This is crucial for reducing the sim-to-real gap, especially for camera-first perception.For engineers, this pipeline is scalable and automated. RealityScan exports critical registration and pose data while supporting standard computer vision (CV) formats (such as Colmap). Its state of the art (SOTA) image alignment provides the backbone for creating Gaussian splats—high-fidelity, photorealistic assets that can be produced with relatively low effort.
RealityScan can run on Linux servers and be controlled by the Remote Command Plugin or CLI scripts, producing automated photogrammetry pipelines that process scans continuously across multiple machines and send the results directly into Unreal Engine.
Synthetic data for fast and cost-effective training
Because Unreal Engine generates every pixel mathematically, it already knows everything about the scene, where every object is, how far away it is, what material it has, and how it’s moving.So instead of requiring humans to label images after they’re captured, the engine can automatically output highly accurate labels at the same time it renders the image.
That has a number of advantages: no disagreements between annotators; no inconsistent labels in a dataset (for example, a forklift labeled as a truck); and no need for an expensive quality-control loop.
What’s more, the economic impact is irrefutable: training simulations on synthetic data significantly reduce the cost per image when performed at scale.
For most perception pipelines, the economic crossover point where synthetic data becomes more cost-effective than manual annotation arrives far earlier than teams expect, often after only tens of thousands of images.
With most robotics models needing millions of images, that means that opting to use synthetic data can result in huge cost savings.
This story is writ large in the market signals we’re seeing—synthetic data is projected to become a dominant component of AI training pipelines.
What’s more, the economic impact is irrefutable: training simulations on synthetic data significantly reduce the cost per image when performed at scale.
For most perception pipelines, the economic crossover point where synthetic data becomes more cost-effective than manual annotation arrives far earlier than teams expect, often after only tens of thousands of images.
With most robotics models needing millions of images, that means that opting to use synthetic data can result in huge cost savings.
This story is writ large in the market signals we’re seeing—synthetic data is projected to become a dominant component of AI training pipelines.
Courtesy of Duality AI
Duality AI leverages these Unreal Engine capabilities for its Falcon digital twin platform, generating high-fidelity synthetic data and sensor simulations for enterprise customers, who leverage them across a variety of industries and use cases.
Investing in these digital twins translates into measurable ROI in terms of rapid training and deployment of AI models to safely fly drones, navigate reliably in the wilderness and even dock orbiting satellites. In each of these scenarios, Falcon’s digital twin environments and virtual sensors deliver highly accurate synthetic data proven to close the sim-to-real gap.
“That would not be possible without Unreal Engine’s advanced rendering capabilities, physics solvers, scalable scene management, and well-established dual use development community,” says Apurva Shah, CEO of Duality AI.
Domain randomization is a technique used to intentionally jumble aspects of a simulated environment so that models trained in simulation generalize better to the real world.
If the model sees enough variation in simulation, it won’t overfit to a narrow synthetic environment—and will work better in messy real-world conditions.
This need for high-fidelity environmental modeling is a priority for Forterra. Their autonomous driving system AutoDrive navigates complex off-road and on-road environments, both structured and unstructured, that lack the predictable markers of a paved city street. From distribution centers and ports to intermodals and natural resource industrial sites, Forterra’s systems must handle varied terrain, shifting surfaces, and unpredictable obstacles in real time.
By leveraging high-fidelity simulation and synthetic data, Forterra can expose its autonomy stack to a near-infinite variety of weather, lighting, and terrain obstacles.
“At Forterra, when it comes to developing autonomy that operates reliably in complex environments, high-fidelity simulation is non-negotiable,” explains Dr. Jeremy Joseph, Forterra’s Director of Simulation & Systems Validation.
“We use Unreal Engine to test our vehicle because it allows us to test in accurate digital environments with physics-accurate sensor simulation and full mission profiles running across an infinite number of weather, terrain, and obstacle variations. This is critical because end-user safety is a core value for Forterra, and these capabilities enable us to identify and resolve issues before they ever reach a real vehicle.”
Investing in these digital twins translates into measurable ROI in terms of rapid training and deployment of AI models to safely fly drones, navigate reliably in the wilderness and even dock orbiting satellites. In each of these scenarios, Falcon’s digital twin environments and virtual sensors deliver highly accurate synthetic data proven to close the sim-to-real gap.
“That would not be possible without Unreal Engine’s advanced rendering capabilities, physics solvers, scalable scene management, and well-established dual use development community,” says Apurva Shah, CEO of Duality AI.
Domain randomization: Preparing for the unexpected
Domain randomization is a technique used to intentionally jumble aspects of a simulated environment so that models trained in simulation generalize better to the real world.If the model sees enough variation in simulation, it won’t overfit to a narrow synthetic environment—and will work better in messy real-world conditions.
This need for high-fidelity environmental modeling is a priority for Forterra. Their autonomous driving system AutoDrive navigates complex off-road and on-road environments, both structured and unstructured, that lack the predictable markers of a paved city street. From distribution centers and ports to intermodals and natural resource industrial sites, Forterra’s systems must handle varied terrain, shifting surfaces, and unpredictable obstacles in real time.
By leveraging high-fidelity simulation and synthetic data, Forterra can expose its autonomy stack to a near-infinite variety of weather, lighting, and terrain obstacles.
“At Forterra, when it comes to developing autonomy that operates reliably in complex environments, high-fidelity simulation is non-negotiable,” explains Dr. Jeremy Joseph, Forterra’s Director of Simulation & Systems Validation.
“We use Unreal Engine to test our vehicle because it allows us to test in accurate digital environments with physics-accurate sensor simulation and full mission profiles running across an infinite number of weather, terrain, and obstacle variations. This is critical because end-user safety is a core value for Forterra, and these capabilities enable us to identify and resolve issues before they ever reach a real vehicle.”
Courtesy of Forterra
Unreal Engine is particularly well-suited for domain randomization because of its programmable material system and dynamic lighting—it’s possible to randomize textures and materials; lighting conditions; weather and atmosphere; object placement and geometry; and camera position and lens parameters.
Robotics teams can also use the engine’s Procedural Content Generation (PCG) framework to vary indoor environments across training episodes. In a warehouse or office scene, you could randomize layouts (for example, shelf or table placement), clutter, like boxes or tools, and visual factors, such as materials, lighting, and occlusions.
And for those looking to construct simulation environments or populate them with objects, the Fab marketplace is a great source for high-quality, affordable, easy-to-integrate 3D content—including highly accurate Quixel Megascans produced using photogrammetry.
Unreal Engine’s rendering pipeline is uniquely suited to simulating the sensors robots use in the real world, including RGB/HDR cameras, depth cameras, LiDAR point clouds, radar and SAR payloads, thermal/FLIR imaging, IMU and GPS models, and underwater sonar (via HoloOcean). It can also model key sensor characteristics such as noise, distortion, latency, and synchronization to better match real hardware.
Duality AI’s Falcon digital twin platform is one example. Its configurable library of simulation-ready sensors includes cameras, LiDAR, radar and SAR, GPS, IMU, and more, with Unreal Engine processing the data to create highly accurate environment simulations.
“We built Falcon with Unreal Engine at its core specifically because we knew how much it has to offer to roboticists and physical AI engineers,” says Apurva Shah, CEO of Duality AI.
Robotics teams can also use the engine’s Procedural Content Generation (PCG) framework to vary indoor environments across training episodes. In a warehouse or office scene, you could randomize layouts (for example, shelf or table placement), clutter, like boxes or tools, and visual factors, such as materials, lighting, and occlusions.
And for those looking to construct simulation environments or populate them with objects, the Fab marketplace is a great source for high-quality, affordable, easy-to-integrate 3D content—including highly accurate Quixel Megascans produced using photogrammetry.
Sensor simulation: Building a comprehensive picture of the world
Modern robotics depends on multi-modal perception, understanding an environment through multiple sensor types at once rather than relying on a single input.Unreal Engine’s rendering pipeline is uniquely suited to simulating the sensors robots use in the real world, including RGB/HDR cameras, depth cameras, LiDAR point clouds, radar and SAR payloads, thermal/FLIR imaging, IMU and GPS models, and underwater sonar (via HoloOcean). It can also model key sensor characteristics such as noise, distortion, latency, and synchronization to better match real hardware.
Duality AI’s Falcon digital twin platform is one example. Its configurable library of simulation-ready sensors includes cameras, LiDAR, radar and SAR, GPS, IMU, and more, with Unreal Engine processing the data to create highly accurate environment simulations.
“We built Falcon with Unreal Engine at its core specifically because we knew how much it has to offer to roboticists and physical AI engineers,” says Apurva Shah, CEO of Duality AI.
Courtesy of Duality AI
dSPACE’s AURELION further shows how Unreal Engine fits into established autonomous vehicle and robotics workflows. For example, the Korean Testing Laboratory (KTL) uses AURELION to test autonomous service robots in a virtual hospital environment, including corridors, elevators, and patient rooms. Virtual sensors enable developers to evaluate navigation, obstacle avoidance, and system response before real-world deployment.
“Unreal Engine is the technological backbone of our physics-based sensor simulation solution, AURELION,” explains Caius Seiger, Product Manager, Sensor Simulation at dSPACE.
“Unreal Engine is the technological backbone of our physics-based sensor simulation solution, AURELION,” explains Caius Seiger, Product Manager, Sensor Simulation at dSPACE.
Courtesy of dSPACE GmbH
The engine enables sensor-realistic camera simulation, including accurate lens and image sensor modeling. “Access to the source code has enabled us to further expand the platform by implementing real-time ray tracing to generate accurate raw sensor data for radar, LiDAR, and ultrasonic sensors,” says Seiger.
Unreal Engine’s efficient integration of 3D models and materials supports a wide range of applications, from urban delivery drones to hospital service robots.
Likewise, MATLAB and Simulink from MathWorks integrate with Unreal Engine via Simulink 3D Animation. “Engineers can use Unreal Engine for photorealistic rendering and sensor simulation, while MathWorks’ tools integrate perception, planning, controls, and physics-based modeling into a unified simulation environment,” says Nishan Nekoo, Product Manager of Simulink 3D Animation.
This workflow helps teams move autonomy development and validation earlier in the design cycle, helping teams uncover issues sooner, more safely, and with greater confidence.
Some robotics organizations build their own middleware to meet the demands of production systems. Tempo Simulation was built with that reality in mind. Their open source platform provides a flexible gRPC/Protobuf interface that enables robotics teams to connect Unreal Engine directly to their own autonomy stacks and middleware. Alongside its TempoROS integration, Tempo also provides tools for building realistic simulation environments, sensor models, and complex agent behaviors inside Unreal Engine.
Unreal Engine’s efficient integration of 3D models and materials supports a wide range of applications, from urban delivery drones to hospital service robots.
Likewise, MATLAB and Simulink from MathWorks integrate with Unreal Engine via Simulink 3D Animation. “Engineers can use Unreal Engine for photorealistic rendering and sensor simulation, while MathWorks’ tools integrate perception, planning, controls, and physics-based modeling into a unified simulation environment,” says Nishan Nekoo, Product Manager of Simulink 3D Animation.
This workflow helps teams move autonomy development and validation earlier in the design cycle, helping teams uncover issues sooner, more safely, and with greater confidence.
ROS integration: Bridging game engine and robot
Compatibility with robotics middleware is sometimes cited as a concern for robotics teams thinking about using game engines. Today, Unreal Engine integrates cleanly with ROS and ROS2 through multiple mature bridges, including ROSIntegration, rclUE, UE ROS2 sensor plugins, and CARLA.Some robotics organizations build their own middleware to meet the demands of production systems. Tempo Simulation was built with that reality in mind. Their open source platform provides a flexible gRPC/Protobuf interface that enables robotics teams to connect Unreal Engine directly to their own autonomy stacks and middleware. Alongside its TempoROS integration, Tempo also provides tools for building realistic simulation environments, sensor models, and complex agent behaviors inside Unreal Engine.
Courtesy of Aurum Systems and Tempo Simulation
“At Tempo, we’ve seen firsthand how powerful the simulation flywheel can be. When teams can generate data, test behaviors, and iterate quickly in simulation, development cycles shrink dramatically. Our goal is to make realistic, high-fidelity Unreal Engine simulations accessible to any robotics team,” says Peter Melick, Co-Founder & CEO at Tempo Simulation.
This is a great example of how Unreal Engine can function seamlessly as a real-time simulator for the software systems that enable robots to operate independently.
When it comes to Reinforcement Learning (RL) whereby an agent learns by interacting with an environment and receiving rewards or penalties for its actions—the realism and responsiveness of the simulation environment directly determine how effectively those learned behaviors transfer to the real world.
Frameworks such as Gym-UnrealCV and AirSim successors integrate Unreal Engine with RL pipelines for drone navigation, object tracking, and robotic control.
Unreal Engine’s Chaos Physics system handles rigid-body dynamics, articulated joints, collisions, and vehicle dynamics. Hybrid architectures combining the engine’s rendering with MuJoCo or Bullet for contact modeling are emerging as best practice for robotics-grade dynamics.
And on the scalability front, Linux containers, Docker support, and GPU-accelerated headless rendering enable cloud-scale synthetic data generation and autonomy testing pipelines.
CARLA is the most widely used open source autonomous driving simulator, now running on UE 5.5 and integrating LiDAR, radar, multiple camera types, GPS, semantic segmentation, and native ROS2 connectivity.
Built on UE5, HoloOcean provides multi-agent underwater simulation with DVL, IMU, optical/acoustic modems, and a novel octree-based sonar sensor model—a rare capability among underwater simulators. UE5’s Lumen and Nanite significantly improve the simulator’s underwater visual realism.
And the Duality AI Falcon digital twin simulation platform, built directly on Unreal Engine, has earned a string of industry recognitions, among which are: Fast Company’s World Changing Ideas, as well as Best Product in AI & Data at the Product Awards and the Data Engines & Simulation category at the Robotics and Physical AI Awards. Falcon’s users include NASA-JPL, Honeywell, DARPA, and Procter & Gamble.
PteroSim by PteroLabs is a UAV flight simulator built on UE5 with JSBSim 6-DOF aerodynamics, supporting fixed wing, multirotor, VTOL, and helicopter aircraft with over 100 simultaneous drones. It runs real autopilot firmware in the loop (PX4, ArduPilot) through the same MAVLink protocol and ground station software used on physical drones, enabling seamless sim-to-real validation.
The platform includes atmospheric disturbance modeling (Dryden turbulence, gusts, microbursts) as well as a gRPC Python API for automated test scenarios, and runs significantly faster than real time for ML training and RL workflows. PteroSim is used by DARPA Lift Challenge teams for pre-flight validation of autonomous aircraft.
"We picked Unreal Engine because the visual quality is unmatched and C++ under the hood means we can run real flight dynamics without compromise,” says Alexander Kalmykov, Founder of PteroLabs. “The engine scales to any hardware configuration, and whenever we got stuck, the UE community already had an answer."
This is a great example of how Unreal Engine can function seamlessly as a real-time simulator for the software systems that enable robots to operate independently.
Supporting the full simulation loop
While Unreal Engine’s primary robotics strength is perception and synthetic data, it supports the full simulation loop.When it comes to Reinforcement Learning (RL) whereby an agent learns by interacting with an environment and receiving rewards or penalties for its actions—the realism and responsiveness of the simulation environment directly determine how effectively those learned behaviors transfer to the real world.
Frameworks such as Gym-UnrealCV and AirSim successors integrate Unreal Engine with RL pipelines for drone navigation, object tracking, and robotic control.
Unreal Engine’s Chaos Physics system handles rigid-body dynamics, articulated joints, collisions, and vehicle dynamics. Hybrid architectures combining the engine’s rendering with MuJoCo or Bullet for contact modeling are emerging as best practice for robotics-grade dynamics.
And on the scalability front, Linux containers, Docker support, and GPU-accelerated headless rendering enable cloud-scale synthetic data generation and autonomy testing pipelines.
Powering major robotics simulation platforms
Unreal Engine is not just a rendering tool. It’s the foundation of several major robotics simulation platforms.CARLA is the most widely used open source autonomous driving simulator, now running on UE 5.5 and integrating LiDAR, radar, multiple camera types, GPS, semantic segmentation, and native ROS2 connectivity.
Built on UE5, HoloOcean provides multi-agent underwater simulation with DVL, IMU, optical/acoustic modems, and a novel octree-based sonar sensor model—a rare capability among underwater simulators. UE5’s Lumen and Nanite significantly improve the simulator’s underwater visual realism.
And the Duality AI Falcon digital twin simulation platform, built directly on Unreal Engine, has earned a string of industry recognitions, among which are: Fast Company’s World Changing Ideas, as well as Best Product in AI & Data at the Product Awards and the Data Engines & Simulation category at the Robotics and Physical AI Awards. Falcon’s users include NASA-JPL, Honeywell, DARPA, and Procter & Gamble.
PteroSim by PteroLabs is a UAV flight simulator built on UE5 with JSBSim 6-DOF aerodynamics, supporting fixed wing, multirotor, VTOL, and helicopter aircraft with over 100 simultaneous drones. It runs real autopilot firmware in the loop (PX4, ArduPilot) through the same MAVLink protocol and ground station software used on physical drones, enabling seamless sim-to-real validation.
The platform includes atmospheric disturbance modeling (Dryden turbulence, gusts, microbursts) as well as a gRPC Python API for automated test scenarios, and runs significantly faster than real time for ML training and RL workflows. PteroSim is used by DARPA Lift Challenge teams for pre-flight validation of autonomous aircraft.
"We picked Unreal Engine because the visual quality is unmatched and C++ under the hood means we can run real flight dynamics without compromise,” says Alexander Kalmykov, Founder of PteroLabs. “The engine scales to any hardware configuration, and whenever we got stuck, the UE community already had an answer."
Courtesy of SAS Institute
Industry 4.0
Industry 4.0, a fusion of robotics, automation, and digital technologies, relies on real-time simulation and digital twins to test production changes virtually. This reduces risk and cost by validating automation decisions before implementation.SAS is an Industry 4.0 leader, combining Unreal Engine’s high-fidelity visualization with SAS AI/analytics Viya to create intelligent digital twins for factory automation. A pilot with Georgia-Pacific uses an Unreal Engine-powered SAS digital twin to visualize and optimize AGV movement, quality control, and maintenance planning in a live plant.
Modern facilities are increasingly complex environments where people and robots must coexist, making safety and efficiency critical.
Glynn Newby, Marketing Manager for Manufacturing at SAS, states, “Our strategy brings together photorealistic simulation with AI and advanced analytics to help leaders build richer digital twins and pressure-test decisions before committing to major investments. Synthetic data moves seamlessly between Unreal Engine and SAS. We believe that this approach has the potential to significantly improve productivity, safety, and operational efficiency across facilities worldwide.”
The evolution of a robotics platform
Unreal Engine was not designed as a robotics platform. But the rise of synthetic data, advances in sim-to-real transfer, and the demand for photorealistic sensor simulation have made it one.Because it is a real-time engine built for interactivity, UE5 supports human-in-the-loop simulation, operator training, teleoperation interfaces, XR experiences, and digital twins, all within the same runtime used for AI development.
Teams can prototype machine autonomy, build interactive visualization tools, and experiment with human-machine interaction workflows without switching platforms.
For robotics teams whose models need to see the real world accurately, whether that means detecting objects on a factory floor, navigating an urban intersection, or identifying anatomy during surgery, Unreal Engine has become the most compelling platform to train in simulation and deploy in reality.