Most robotics simulation tools lack realistic context, making it hard to get the data you need for training your machine learning algorithms. Duality’s realistic synthetic environments accelerate machine learning with rich perception data at scale.
We customize virtual worlds with interactive digital twins of your intelligent machines. As you stress-test your latest algorithms in realistic scenarios—such as infrastructure inspection or urban take-off and landing—our solution captures scenario data and exports it in a variety of formats including ROS, CSV, and USD.
The Duality solution leverages the technology and rich application ecosystem of Unreal Engine combined with a flexible Python interface and API that easily integrates with your autonomy stack.
High Performance Simulator for Intelligent Machines
Duality’s Falcon Sim imagery is so realistic that machine learning networks can’t tell the difference between ours and that from the real world. Better imagery means better data for faster learning, bigger scenario libraries, and more intentional edge cases.
See how Duality’s simulation technology helps you validate safe operation of your machines under diverse conditions and gather data significantly faster than with field testing.
We leverage the full power of Unreal Engine, making it possible to craft realistic environments of any type. We model and distribute features in these environments with intentionality to facilitate machine learning training.
Our realistic environments and accurate sensor models from cameras to LIDARs allow us to output segmentation data to help train your machine learning algorithms. Our customers’ systems perform the same using our imagery as they do on their own logs, with equal pixel accuracy and IoU scores. Even better, they fail the same way: Problem cases like hard shadows cause the same failures in the real and virtual worlds.
We can easily add semantic labels to our synthetic environments simply by turning them on. Every element in our environments is intentionally crafted, and each element carries metadata that allows us to surface semantic labels and other analytics data.
Once our multidisciplinary team builds your custom simulation environment, our World Builder feature enables you to easily manipulate a scene to adjust and refine challenging scenarios.
A Fortune 100 company integrated Duality’s solution into its simulation pipeline to generate training data for a UAVs machine learning system. Our flexible solution may be customized to support a wide range of other use cases in other industries.