Gatik AI Inc. today announced Arena, a new simulation platform to accelerate the development and validation of its autonomous vehicles, or AVs. Arena produces structured and controllable synthetic data that addresses the limitations of traditional, real-world data collection, according to the company.
“As the AV industry pushes toward scaled deployments, the bottleneck isn’t just better algorithms — it’s better, smarter data,” stated Gautam Narang, co-founder and CEO of Gatik. “Arena allows us to simulate the edge cases, rare events, and high-risk scenarios that matter most, with photorealism and fidelity that match the complexities of the real world.”
Founded in 2017, Gatik said it is a pioneer in autonomous middle-mile logistics. The company‘s systems have been commercially deployed in Texas, Arkansas, Arizona, and Ontario.
Arena combines AI techniques
Capturing exceptions in real-world AV testing is time-consuming, expensive, and unsafe, Gatik noted. “Traditional fleet testing and data logging cannot provide the scale, diversity, or reproducibility required to validate AV systems comprehensively,” it said.
Arena uses an extensible, modular simulation engine that combines different AI techniques, including neural radiance fields (NeRFs), 3D Gaussian splatting, and diffusion models. It uses volumetric reconstruction to create high-fidelity simulations from abstract representations such as segmentation maps, lidar, and HD maps.
Gatik also said Arena combines real-world logs, trajectory editing, agent modeling, and multi-sensor simulation pipelines to deliver full, closed-loop simulations. It can adjust traffic flow, pedestrians, lighting, and road layouts for scenario editing and A/B testing.
“Arena provides an ecosystem of tools and allows digital simulation to scale up,” said Apeksha Kumavat, co-founder and chief engineer of Gatik. “It can create photorealistic data, and the end-to-end simulator allows us to simulate multiple sensors — cameras, lidar, and radar — as well as vehicle dynamics.”
“Traditionally, simulators have been been based on physics-based game engines, and they could test certain parts of the autonomy stack, but not end to end,” she told The Robot Report. “That took a lot of resources and led to a sim-to-real gap. Now, this simulator reduces that gap to close to zero, and we can do a lot of data collection and synthesis in the ecosystem itself.”
In addition, Arena can reflect real-world behavior of sensors under varied environmental conditions. By simulating interactions between self-driving vehicle decisions and surrounding agents, the platform enables testing of the full autonomy stack in interactive environments. Gatik said this includes modeling vehicle dynamics, policy interactions, and latent scene evolution.
“We can now truly replicate the world in a digital twin, with all the sensor noise and variations,” said Kumavat. “Reducing the sim-to-real gap allows us to have the confidence to use the data for training and true safety validations.”
Synthetic data sufficient for Gatik’s safety case
Arena supports generation of structured synthetic data for machine learning workflows, regression testing, and safety case validation without requiring a lot of annotated real-world data, said the company.
“With Arena, we’re reimagining simulation not just as a testing tool, but as a core enabler of safe, scalable autonomy,” said Narang. “It gives us the control, realism, and flexibility we need to rapidly build confidence in our systems-and do so without compromising safety or time to market.”
Arena is able to model safety-critical scenarios such as bad weather and visibility, unpredictable road users, challenging road geometry, dynamic road changes, sensor occlusions or failures, and dense urban interactions. The goal is scalable, safe, and repeatable AV testing in highly realistic digital worlds, said Gatik.
“We’ve been using Arena for a little while to scale up development, training, and validation,” said Kumavat. “This can go further in terms of expanding scenarios, but it can also translate into different geographies. With diffusion and foundation models, it can adapt to Toronto or Europe, and this ability to change while still grounded in physics allows it to scale.”
Arena enables manipulation of conditions such as weather in AV simulations. Source: Gatik
NVIDIA collaborates toward autonomous freight
For Arena, Gatik has collaborated with NVIDIA to integrate NVIDIA Cosmos world foundation models (WFMs) to create high-fidelity, physics-informed digital environments for robust AV training and validation. The partners announced earlier this year that Gatik will use NVIDIA DRIVE AGX with the DRIVE Thor system-on-a-chip (SoC) to serve as the AI brain for next-generation autonomous trucks.
“NVIDIA Cosmos has been purpose-built to accelerate world model training and accelerate physical AI development for autonomous vehicles,” said Norm Marks, vice president of global automotive at NVIDIA. “Our collaboration with Gatik unlocks the development of safe, reliable, ultra-high-fidelity digital environments for robust AV training and validation, and is helping to accelerate the commercialization of Gatik’s autonomous trucking solution at scale.”
“We’ve been working with NVIDIA for a while on hardware chip sets, and Gatik had been using Orin for a while,” said Kumavat. “We’ve been working with NVIDIA for a year on this particular software for autonomy. We’re able to use these WFMs for a simulation use case adapted to our domain.”
“Simulation is a subset of the whole Arena ecosystem,” she explained. “Edge cases were a key thing gating the application. Safety teams had to manually define boundary conditions themselves or run [actual vehicles for] millions of miles to uncover a few edge cases. It was a resource-intensive process.”
“Now, we have generative AI-based adversarial scenario mining,” Kumavat said. “We can run millions of edge cases more exhaustively to find boundary conditions, making the process easier. Knowing the boundaries of a system affects safety, and we’re working on more exhaustive safety cases that will be validated by third-party auditors and provided to all stakeholders including regulators.”
She acknowledged that Gatik and NVIDIA needed to make sure that there was an architecture for keeping physics grounded in the real world, verifying AI’s output, and aligning onboard and off-board processes. “There are a lot of guardrails to ensure the sanity of data, and we’ve struck a balance between the need for real-world testing and relying on simulated sensors. We’ve created functional metrics for checking how close the simulation is to the real world.”
Gatik asserted that the platform will reduce reliance on road testing and accelerate commercialization of its autonomous trucks for partners including Kroger, Tyson Foods, and Loblaw.
“Today, we have 100 vehicles on the road with different customers, and we expect 10x growth in the coming years,” said Kumavat. “These are not one-off pilots but are multi-year contracts. We’ve already realized a lot of value from using frameworks like Arena for customers that are already deployed, but it allows us to expand in existing geographies and with new customers.”
Tell Us What You Think!