With our naturalistic traffic flow simulation it is possible to test the software for automated driving functions on the basis of simulated endurance test runs. The tests performed help to uncover unforeseen edge cases and provide information about the robustness of the implemented algorithms of our customers.
AAI traffic agents are modelled with a focus on vehicle-to-vehicle interaction and our special mix of stochastic, supervised learning and reinforcement learning. Different aggressiveness levels, vehicle mixes, traffic densitis and cultural backgrounds ensure a naturalistic traffic pattern and more or less challenging situations for the test vehicle - without parametrizing effort for simulation engineers.
✓ The most realistic way for simulating endurance test drives on highways and in cities
✓ Traffic agents with independent decision-making based on tunable levels of aggression
✓ Culture-based driving behavior through training with real traffic data
✓ Tunable vehicle mixes and traffic densities
✓ No time intensive and engineer dependend programming of traffic agents
✓ Large selection of maps + every map in OpenDRIVE format is integrable
✓ Possibility to make individual changes in the loaded map manually or even to create an own synthetic map
✓ Accurate driving logs for deterministic replay and analysis
✓ Possibility to extract critical or interesting scenarios automatically in OpenSCENARIO format via AAI Scenario Extraction feature
It is a complicated journey for our traffic agents to learn human-like behaviour. Like student drivers, our driver models must first learn the basics of driving and the applicable traffic rules. But the journey does not end here, because in the real world, much more complex behavior can be observed than the model behavior taught in driving schools. Due to unwritten rules and local habits, negotiation between road users is often necessary.
Therefore just like ourselves, the AAI traffic agents never stop learning and are being continuously trained using real-world traffic data, using supervised and reinforcement machine learning methods.
The benefit of simulated endurance testing can only be fully exploited with naturally acting traffic around the ego vehicle.
Our traffic agents support a range of driver profiles, e.g. more conservative and defensive ones, who prefer to obey traffic rules and drive at constant speed, or very dynamic and even aggressive ones, who tend to break the law occasionally and assert themselves with lots of lane changes, more extreme acceleration and speed.
All incidents are integrally documented and can be replayed for analysis with AAI Replicar. Interesting maneuvers from simulated endurance tests can be transscribed into deterministic scenario templates and stored in the scenario database for repeated testing.
AAI Intelligent Traffic is a stand-alone and highly compatible software module that can be integrated into various simulation environments. Supported interfaces include Functional Mock-up Unit (FMU/FMI), specific middleware and much more.
Let us know about the setup of your simulation environment and we will help to integrate the traffic module according to your requirements.