Cooperative Autonomous Driving with Safety Guarantees (KoSi)

Statement

Benefits of automated driving will only be fully realized if vehicles behave cooperatively. A major obstacle to this promising vision is a mathematically based safety guarantee that safeguards the automated system under any environmental conditions. Efforts in this direction are particularly challenging because every driving situation is different and thus the driving functions to be automated cannot be verified in advance, so the vehicle must constantly verify itself. In addition, it is not yet clear how artificial intelligence for autonomous driving functions can be efficiently safeguarded.

Main Objective

We will realize comfortable, economical, and safe driving with a unified approach by progressively refining safe and rapidly predictable solutions using artificial intelligence. A novel approach will be developed in which automated vehicles cooperatively coordinate driving strategies that are subsequently optimized and verified. The necessary calculations are distributed across multiple communicating vehicles, while still guaranteeing correctness - even in case of communication failure. Robust sensor technology taking AI elements into account is used to prevent manipulation by possible attacks.

Implementation

We create a framework that holistically considers driving behavior prediction, maneuver and trajectory planning, and formal verification to guarantee inherent correctness of vehicle guidance. Specifically, we consider correct behavior under communication failure, mixed traffic with human drivers and automated vehicles, parameter uncertainties, sensory capabilities, external influences, and the future behavior of other road users.

Developed Tools and Benchmarks

• commonroad.in.tum.de
• MINIVAN (Mixed INteger InteractiVe plAnNing)

Publications