aiSim is an integral part of AImotive’s development efforts that runs on a purpose-built engine. Unique design solutions make aiSim especially effective in training neural networks, allowing developers to test self-driving software, such as aiDrive, in any virtual scenario. aiSim's flexibility drastically increases the amount of data available to train artificial intelligence algorithms, and the time self-driving cars spend "on the road."

Data Generation & NN Training

Utilizing the scalability of computer systems, aiSim can create hundreds of thousands of image sets a night, running on only a handful of developer PCs. The ability to export virtual data for any of the cars sensors drastically increases the amount of data available to recognition, localization and motion planning algorithms. aiSim can generate image sets to validate sensor calibrations, while supersensor data can also be generated for decision making training.

Scenario Testing

The standalone scenario editing tool offers the ability to create scenarios and define goals, set constraints and traffic behaviors in various situations. The simulation environment allows fully customizable physics implementation and a unique sensor setup for the self-driving vehicle. Besides locations, weather patterns, and other variables, hundreds of corner cases can be generated that are extremely rare in real-world testing. To expedite scenario creation these small but drastic changes lead to thousands of variants based on a single scenario.

Modular Verification

aiSim allows modular verification for individual components of autonomous software. For example, motion planning can be verified through testing single decision-making algorithms with a high-level overview of the situation, allowing for a scalable testing process. Testing recognition functionality is possible utilizing the simulator’s ability to recreate real-world environments, virtual sensors, and sensor failures or degradation. These possibilities make aiSim a safety-aware development platform.