Intersection Management

Print

The Concept of Decentralised, Autonomous Intersection Management

A combination of drive-by-wire technologies and communication between different vehicles (car-to-car, car-to-infrastructure) makes possible a large number of autonomous driving functions.

Vision of an intersection without traffic lights

So a scenario involving vehicles that autonomously cross an intersection without traffic lights comes into reach and promises several advantages:

  • increase in road safety
  • increase in intersection capacity
  • reduction of waiting time for road users
  • reduction of fuel consumption of the vehicles

The autonomous intersection management deals with this scenario resp. with the computation of optimal trajectories for the vehicles in the intersection. It takes into account not only collision avoidance but also other objectives such as crossing time, comfort, or fuel consumption.

  • Simulation of an intersection
  • Simulation of an intersection

History

Initially the intersection management served as an application example in the context of the Collaborative Research Center 376 (Massive Parallelism) because at that time it was assumed that the complex task of an almost simultaneous optimisation of a large number of vehicles in an intersection area would be impossible to perform for a single computer under real-time conditions. For this reason all vehicles in the intersection area are to pool the unused computing capacities of their ECUs in a heterogeneous and powerful processor network so as to help solve the problem of all vehicles autonomously crossing the intersection in due time in a massively parallelised manner.

The Solution Algorithm

In the context of the CRC 376 several procedures for an intersection management have been developed. The latest, most efficient procedure consists in formulating the task of determining an optimal trajectory for a vehicle in the shape of a network-flow problem.

The prerequisite is a discretization of the processes in the intersection area. The algorithm developed for solving the problem is mainly based on Dynamic Programming according to BELLMAN resp. on the BELLMAN-FORD algorithm for finding so-called shortest paths. The trajectories of the vehicles in the intersection area are computed sequentially. Here, the maximum number of required computations per trajectory or rather the worst-case runtime of the algorithm is known on the one hand while on the other hand, the scaling can be tailored to the respective performance of different target systems. Thus parallelised computations are no longer compulsory even if parallelisation would make the procedure more efficient. Contemporary off-the-shelf computers, e.g., a single-core PC with 3 GHz, can compute a trajectory in less than 50 msec, depending on the discretization and the configuration of the optimisation. Consequently, it is immediately evident that the algorithm is suitable for real-time applications – a feature that was nonexistent in earlier procedures.

Simulation and Results

The procedure developed was implemented in a simulator that is able to simulate up to 4 traffic junctions (intersection and roundabout) simultaneously on different computers. A characteristic trait of an intersection is that its lanes are subject to mixed use, i.e., there are no separate turning lanes but all vehicles, whether running straight ahead or turning left or right, share the same entrance lanes. As yet, trucks and pedestrians have been left out of the simulations. In these circumstances the procedure can increase the capacity by a factor of 3 to 4 as set against a comparable intersection with a conventional traffic-light system.

Results of a simulation run

Downloads