Autonomous driving

As a key emerging technology, autonomous (or machine) driving (AD) represents a key applied research area of our chair. Developing AD solutions requires a mindset of mechatronics, which goes far beyond its interpretation in mechanical engineering. Perspectives of control engineering, computer science and communication engineering offer versatility and dominate the mechanical domain as solutions and use-cases demand higher levels of autonomy and connected driving. Our AD solutions to-date conjunct adaptive control algorithms with various modules for environment perception. Control algorithms are developed using not only the well established concepts such as model predictive control (MPC), but also the modern machine learning techniques such as IL, RL and DNN based stochastic MPC. The developed algorithms target various driving scenarios, including city driving, inter-city and highway driving, valet parking and more. The developed algorithms are tested in real-world scenarios using two AD facilitated vehicles, which are equipped with high-end sensory and computation resources.
 

Goals

Our vision is to develop an end to end algorithm that is able to perform all three core tasks of perception, planning and control. Some of the key research topics are:

  • Development of a unified mulit-sensor perception module: The goal here is to extend the vision based perception module to incorporate surround 3D information based on LiDAR data.
  • Development of a robust decision-making and planning module: In order to deal with highly diverse traffic scenarios our planning module is based on both 2D and 3D information. The former, which is based only on camera images, is responsible for proposing different plausible maneuvers along with a target (virtual) drivable path for the ego vehicle. Alternatively, different such target paths for the ego-vehicle is also made available in the vehicle-coordinates.
  • Development of a robust vehicle controller: At the heart of our AD algorithms lies the vehicle-controller which is responsible for, safely and reliably, steering the vehicle through traffic. Our control algorithms are developed using not only well established concepts of MPC and robust control but also using modern machine learning methods such IL, RL and DNN based stochastic MPC. In order to facilitate seamless communication with DNN-based planning and perception modules, especially for the case of a purely camera based solution, our recent efforts are more in the direction of developing a direct vision based DNN controller.
     

Video

References

Learning based interpretable end-to-end control using camera images
20th International Conference on Informatics in Control, Automation and Robotics (INCINCO), 2023. DOI
S. A. Hiremath, P. K. Gummadi, A.Tika,  P. Rama and N. Bajcinca

DNN based Learning Algorithm for State Constrained Stochastic Control of a 2D Cartpole System
European Control Conference, London, 2022
S. A. Hiremath and N. Bajcinca

NIAR: Interaction-aware Maneuver Prediction using Graph Neural Networks and Recurrent Neural Networks for Autonomous Driving
IEEE International Conference on Robotic Computing, 2022
P. Rama and N. Bajcinca

Distributed optimization for feedforward global chassis control
IFAC Advances in Automotive Control AAC, Munich, Germany, 2010
N. Bajcinca and Y. Kouhi

Autonomous all-wheel car steering
IEEE International Conference on Control Applications, Munich, Germany, October 4-6, 2006 DOI
N. Bajcinca

Keywords

Autonomous Driving
Environment Perception
Decision Making
Vehicle Control
 

Contact

Dr. Sandesh Hiremath
Gottlieb-Daimler-Str. 42
67663, Kaiserslautern
Phone: +49 (0)631/205-3455
Fax: +49 (0)631/205-4201
sandesh.hiremath(at)mv.uni-kl.de