My ADAS team at Siemens Digital Industries Software works on the intersection of simulation, physical testing, and algorithms. We have been trying to find ways to leverage logged data futhur and more efficiently serving those objectives. Here are some of our approaches to make the ADAS dataset more active and lively, including interesting demos from our R&D works.
-
Real2sim: import real dataset (i.e. from Waymo Open) automatically into simulation (Simcenter Prescan, Amesim), where sensor, vehicle models, and traffic scenarios can be tuned and analyzed.
-
Adapt to diverse traffic driving styles: just imagine a scenario collected in US but some vehicles now drive with Japanese driver style…
-
Closed-loop testing: replace ego vehicle in the dataset by your ADAS controller - MPC or AI (imitation learning, reinforcement learning)
-
Put human on driver seat: reply data on driving simulator together with digital twin real-time models, enable to evaluate perceived safety and comfort.
-
Generative AI: to synthesize new traffic scenarios.
Enjoy and feel free to drop your comments :) !