The Simulation Engineer Moving from Game Engines to Robot Training Grounds

Game and robotics companies have begun looking for the same expertise

Krafton and the Interdisciplinary Program in Artificial Intelligence at Seoul National University's Graduate School held a joint research launch ceremony at the university on October 8. They announced that they would pursue specific research projects in world models over the next 3 years[1]. World models let AI understand its current situation, then predict and simulate how its surroundings will change after an action. The two institutions will study prediction and decision-making technologies needed both in virtual game environments and in the real environments where robots operate.

There are three research projects. The first develops world models that keep spaces and objects visually consistent even in long videos. The second uses motion data from games to connect the movements of people and surrounding objects naturally. The third uses virtual scenes to test whether AI actually bases its decisions on spatial information such as passage widths and distances to objects. Each project will also develop benchmarks to assess its results[1]. Krafton established a robotics research company in the United States last year and also established a Korean entity in February this year. Krafton's chief AI officer (CAIO), Lee Kang-wook, said, "The capabilities built through games in simulation and virtual-world creation are an important foundation for world-model research"[1].

The robotics side is looking for the same capabilities. Maum AI's WoRV team is recruiting a simulation engineer from October 7 to 31[2]. The posting's "Who we are looking for" section says, "People who want to extend the engineering skills they have built in games and graphics to robotics and Physical AI." Six days later, on October 13, the Ministry of Trade and Industry's Industrial Convergence Regulatory Sandbox Committee approved an exemption allowing Maum AI to conduct demonstration operations of autonomous quadruped patrol robots on riverside walking paths and in parks around Cheonan[3].

At the same meeting, Robros received a demonstration exemption to deploy 20 AI-equipped bipedal humanoids at sites in semiconductors (Nepes), automotive manufacturing (Sungwoo Hitech), shipbuilding (HD Hyundai Samho) and logistics (Delivery Lab)[3]. Until now, industrial robot safety requirements assumed fixed installations surrounded by safety fences. This prevented humanoids that move and work alongside people from entering these sites. The government allowed this demonstration subject to a risk assessment and safety measures required by the Ministry of Employment and Labor.

Engine experience is required, robotics experience is preferred

The four lines of qualifications in Maum AI's posting do not mention the word "robot"[2]. They call for the ability to develop and optimize large-scale software in C++ and Python; experience developing or extending physics engines, rendering engines, game engines or 3D simulators; a deep understanding of 3D mathematics and rigid-body dynamics or rendering pipelines; and the ability to quickly understand unfamiliar codebases and make trade-offs between accuracy, performance and maintainability.

Robotics experience is preferred. This includes experience integrating ROS and ROS 2, transferring policies learned in simulation to real robots through Sim2Real, and building environments to evaluate reinforcement learning (RL) or vision-language-action (VLA) models. The other preferred qualifications return to engine work: modifying physics engines such as PhysX or MuJoCo or implementing deformable-body simulations such as FEM and MPM; programming GPUs with CUDA or Vulkan and optimizing real-time rendering; and developing Isaac Sim and Omniverse Kit extensions and designing large-scale scenes based on USD[2].

The order is reversed compared with the usual route into robotics engineering. Across specializations, robotics engineering jobs commonly require experience in control, robot kinematics, computer vision and robot middleware such as ROS. This posting asks first for engine expertise built in games and graphics, with robotics knowledge listed as a preference.

Building environments, generating data and measuring the gap with reality

Reading the posting's six lines of main responsibilities in sequence reveals the scope of the role[2]. The first stage is environment building. Engineers create realistic simulation environments in Isaac Sim and produce 3D assets ready for use in simulation (SimReady asset). They also develop asset pipelines connected to digital content creation tools (DCC), along with internal tools. They implement simulations of phenomena that the physics engine does not handle by default. The posting gives accumulated snow as an example of a material that requires continuum mechanics calculations.

Next comes large-scale data generation. Engineers reduce bottlenecks in physics calculations, rendering, sensing and data serialization. They also build infrastructure that runs multiple simulations simultaneously without a display (headless) to generate large datasets. They create low-latency communication links between the simulator, robot software and AI models.

The final stage is validation. Engineers model robots, apply domain randomization to deliberately vary conditions such as lighting and materials, and create benchmarks. This helps assess whether behaviors learned in simulation also work on real robots[2]. Krafton and Seoul National University's research projects will each develop benchmarks as well[1]. Measuring how much a virtual environment differs from reality is as much a part of the job as building that environment well.

Game developers, graduate students and robotics engineers take different routes in

For game and graphics engineers, the tools they already use provide an entry point. Physics engines, rendering pipelines, GPU optimization and shaders appear across the required and preferred qualifications in Maum AI's posting[2]. Starting with simulation environments and assets, moving into infrastructure for large-scale data generation, and then expanding into Sim2Real evaluation and benchmark design matches the scope of work in the posting. ROS is a preferred qualification. For someone preparing independently, connecting a small robot simulation to ROS 2 can fill a gap in their application.

Graduate students have a route through research. Krafton and Seoul National University agreed to hold research exchange meetings and run student internships at their joint research center, as well as identify new joint research projects[1]. The launch announcement does not specify when internships will recruit or how many students they will take. As of October 16, Krafton's AI Research division job board lists a 6-month research engineer internship studying VLA agents that take screen content, text and game state as inputs and produce actions in game environments. Applicants must be pursuing a master's or doctoral degree in artificial intelligence, computer science or a related technical field[4].

Robotics engineers enter from the other side. For those who understand control and kinematics and have integrated ROS 2, that experience maps directly onto the preferred qualifications. The gap to fill is engine expertise. To identify differences between simulation and real robots, they need to explain how physics engines calculate contact and friction, and how rendering settings affect camera sensor data. Applications for Maum AI's role close on October 31[2].

References

[1] Electronic Times. Krafton and Seoul National University launch joint world-model research, expanding AI beyond games into robotics (in Korean). 2026-10-08.

[2] Linkareer. Maum AI WoRV team Simulation Engineer recruitment (in Korean). Applications 2026-10-07~2026-10-31, checked 2026-10-16.

[3] Yonhap News Agency. Report on humanoid and patrol robot demonstration exemptions at the Ministry of Trade and Industry's 6th Industrial Convergence Regulatory Sandbox Committee meeting (in Korean). 2026-10-13.

[4] Krafton. AI Research Division Research Engineer, Vision Language Action Game Agent job posting (in Korean). Checked 2026-10-16.