Technology 

Highwai provides customizable simulated worlds for training autonomous vehicle AI. Our safe and custom worlds help overcome real world limitations in edge use-case scenarios. A combination of detailed scene environments, scenario building, ground truth annotations and reinforcement learning makes Highwai the ideal simulator for autonomous vehicle AI. 

 
 
scenario_manager_image.PNG

ScenARIO editor

Place moving objects, movement paths and behaviors within the environment. Scenario Editor allows the flexibility to set movement paths for pedestrians and vehicles, set velocities, change time of day, visibility and weather. Includes physics-accurate shadows, wind motion and lighting in order to create the most lifelike test environment possible. Choose from millions of auto-generated, repeatable training scenarios or create your own. 

 
simulator_peds_cars_image.PNG

scenE editor

Generate any scene using static objects in a simulated environment. Use objects such as signs, buildings, trees, traffic lights, and hills to create an extensive, detailed city model ideal for AI interaction and learning. Features a scene library and asset library to help streamline the scene creation process.

 
60 degree false color flash LIDAR distance - ground truth.PNG

AI toolset

The AI Toolset provides a complete deep learning workflow including dataset management, data augmentation, and neural network training, inference and regression. The toolset also includes data visualization tools and statistics analysis packages.

 
Reinforcement_Learning_website_image.jpg

REINFORCEMENT LEARNING

Reinforcement learning is at the core of artificial intelligence and plays a key role in autonomous vehicle technology. Data output is collected from the simulator, plugged into and cycled back through the simulator. This reinforcement learning cycle allows for more accurate and improved AI performance.

 
Device_Testing_website_image.jpg

device testing

Plug in hardware or software into the simulator's pluggable subject car for device testing. Run the subject car through the simulator, then compare objects detected against 100% accurate ground truth. Training data is color annotated to create a concise, visual way of viewing data.