Automated Driving Systems Testing Using Agent-Based Modeling

Southwest Research Institute (SwRI) has gained worldwide attention by leading NASA missions such as the New Horizons mission to Pluto and the Juno mission to Jupiter. SwRI is also a leader in fuel and energy efficiency, geosciences, turbomachinery, and energy storage. Their contract engineering efforts benefit government, industry, and the public through the application of science and technology.


One of the institute’s research areas is automated driving systems. SwRI has been working in this field since 2006 and has designed systems for a semi-truck, Ford Explorer, many military platforms, and a wide variety of unmanned aerial vehicles (UAVs), commonly known as drones. These automated systems no longer need human drivers to control them on their missions, whether it is reconnaissance, hauling, or simply transportation.

But SwRI’s engineers didn’t want to stop there and decided to make autonomous vehicles free, not only from the driver, but also from a control center. According to this idea, vehicles would communicate in a distributed manner with each other, share information about their current location and environment, and make decisions on further actions based on this information themselves. This technology would primarily be used by military forces for the transportation of supplies to the fields of operation, demining, reconnaissance operations, and many other areas where humans can be replaced by machines for their own safety.

The implementation of such systems can take a lot of time and money, so the engineers at SwRI decided to use simulation modeling to explore the possibilities of autonomous vehicles.


Collaborative map of the area based on agents’ explorations

Picture 1. Collaborative map of the area based on agents’ explorations

To assess the performance of automated vehicles and to evaluate algorithms and task sharing between the vehicles, SwRI engineers decided to build an agent-based AnyLogic model of vehicles’ operations in an enclosed area with random obstacles. It was the easiest way to represent multiple interacting virtual vehicles with a variety of capabilities and have them all operate simultaneously.

The vehicles detected obstacles, found, and refueled capsules in the area. Completing these tasks quickly required the vehicles to cooperate and share information about the environment.

All the vehicles had sensors that could detect the environment, gather information about the things around them, and share their knowledge with other agents. Each vehicle was given predefined behavioral features: some vehicles could only search for capsules, some could only check if capsules were full or empty, and others could only refuel them.

Agent’s state chart

Picture 2. Agent’s state chart

On the right side of Picture 1, you can see how each individual vehicle explored the area and found obstacles, and on the left side of the picture is the collaborative map of the area based on their explorations. The collaborative map was shared between all the agents and each agent could take advantage of the combined mapping capabilities of the other ones.

For refueling capsules in this area, vehicles had to form teams based on their individual capabilities and location. When a searcher vehicle finds a capsule, it signals for the nearest vehicles with the required capabilities. These vehicles create a team with the searcher and assist with classification and refueling to complete the mission. In Picture 2 you can see the agent’s state chart describing this process.


With AnyLogic, SwRI engineers tested how automated vehicles could behave in a cooperative network, and proved that such networks can be built in real life. Consequently, this suggests that researchers can develop algorithms for solving any related problem using modeling in AnyLogic, test these algorithms, and implement them in autonomous vehicles. For example, the creation of a mixture of drone and ground-based robots for scouting or security patrols.

To learn more, watch the project presentation at AnyLogic Conference 2016 or download it.

Ähnliche Fallstudien

Mehr Fallstudien