Overview

Autonomous driving has attracted a lot of attention during the last years and faces many technical challenges regarding online control and object detection in sensor data. Our focus is on the control part of autonomous or automated vehicles and we are developing optimization-based path planning and path following techniques. This includes model-predictive control schemes and machine learning approaches. A particular focus is on the coordination of interacting vehicles. To this end we are creating a framework which combines real and virtual systems. This framework allows to investigate even potentially dangerous situations in a safe and reproducable way. Moreover, it is flexible in the sense that human drivers and virtual or real automated vehicles can interact in realistic traffic scenarios. 

coupling_virtual_real.png

 

The following methods are

  • real-time optimization methods and sensitivity analysis for path planning
  • linear and nonlinear model-predictive control (MPC) for path following and path planning
  • dynamic inversion control for path following
  • DAE control techniques for path following
  • coordination of interacting vehicles using hierarchies, generalized Nash equilibria, and dynamic scheduling 
  • reinforcement learning (RL) for path planning and collision avoidance
  • dynamic programming (DP) for path planning and dynamic scheduling
  • reachable sets for worst-case prediction and probability-based prediction

 

The methods are implemented and tested on our test vehicles:

 

We use the robot operating system ROS2 and the Unreal Engine for visualization.

Autonomous car driving on testtrack

The video shows the drive of an autonomous car on the testtrack of the University of the Bundeswehr Munich. The path following and path planning tools with collision avoidance capability have been developed within the dtec.bw project MORE by the Engineering Mathematics Group (LRT 1.1) of the Department of Aerospace Engineering.

GNEP-MPC for Coordination of Interacting Vehicles

The video presents a control experiment for interacting vehicles in a road network. The approach combines a high-level controller for the generation of collision-free trajectories and a low-level dynamic inversion controller for path tracking. The high-level controller uses model-predictive control for generalized Nash equilibrium problems, which are used to coordinate the vehicles. The control concept was implemented and validated on scale robots.

Online NMPC with obstacle avoidance

The video shows the online control of a scale car with nonlinear model-predictive control (NMPC) and obstacle avoidance. The position is obtained by an indoor GPS system. The online optimization within the NMPC is done by the software OCPID-DAE1.

Automated driving using a finite state machine

The video illustrates a finite state machine approach for automated driving developed by Mostafa Emam from the Engineering Mathematics Group of the Universität der Bundeswehr München. The corresponding paper was presented at VEHITS 2022. The approach is simulated in the 3D visualization environment ROCS with the aid of Omid Moslehirad.

Intersection management using dynamic scheduling

The animation shows examples of a dynamic scheduling algorithm for automatic intersection management for autonomous cars. To this end a scheduling problems is coupled with optimal control subject to collision avoidance.

ROCS - Realtime Optimization and Control Simulator

The Realtime Optimization and Control Simulator (ROCS) is a software package written in Qt3D with Unreal Engine frontend. It is being developped at the Engineering Mathematics Group of the Universität der Bundeswehr München. It is a versatile tool to visualize and control vehicles, aircrafts, and robots in complex scenarios.

Distributed Model-Predictive Control with Car-to-Car Communication

Modern vehicles offer a number of passive and active driver assistance systems, which shall support the driver in crucial situations. The degree of automation of such systems increases continuously and eventually ends in systems that act fully autonomously within given bounds. Such assistance systems get more and more important in controlling cars in all-day road traffic. In future cars will communicate and negotiate driving strategies. The following video illustrates a distributed model-predictive control algorithm for autonomous cars in typical traffic scenarios using car-to-car communication.

Hierarchical Control of autonomous connected vehicles

Data like position, velocity and heading are transmitted and shared among the vehicles. Regarding these informations a unique hierarchy level for each vehicle is derived according to predefined rules. Superordinate vehicles have to be considered for collision avoidance by vehicles with lower priority. Whereas vehicles with lower priority are neglected.

Automatic testdrive along a sloped road

The animation shows the automatic testdrive along a sloped road. The mathematical model is based on the single track model, which was augmented by dynamic tyre loads and force terms taking into account the longitudinal and lateral slope of the road. The trajectory was computed using OCPID-DAE1 using a model-predictive control scheme.