• Multiobjective discovery of driving strategies for autonomous vehicles
The Client : Javna agencija za raziskovalno dejavnost Republike Slovenije ( Z2-7581 )
Project type: Research projects ARRS
Project duration: 2016 - 2018
  • Description

There  have  been  several  investments  in  the  autonomous  vehicle  driving  technologies  by automotive  companies  recently.  The  focus  of  these  technologies  is  mostly  on  determining  the vehicle  surroundings  in  order  to  increase  the  driving  safety  and  avoid  collisions.  In  addition, several  researches  have  developed  optimization  algorithms  for  discovering  driving  strategies, which  optimize  also  other  driving  aspects  such  as  the  traveling  time,  fuel  consumption, emissions etc.


The proposed project aims at developing an algorithm for discovering driving strategies that, on the one hand, could be deployed in real vehicles, and, on the other hand, would be also suitable for  solving  hard  optimization  problems  such  as  real-time  discovery  of  driving  strategies  with short traveling time and low fuel consumption.


The  leader  of  the  proposed project has  already  developed  a  Multiobjective  Optimization algorithm  for  discovering  Driving  Strategies  (MODS) that  optimizes  the  traveling  time,  fuel consumption  and  driving  comfort.  The  results  show  that  MODS  finds  better  driving  strategies than  existing  algorithms  for  discovering  driving  strategies.


During  the  proposed  project,  the  MODS  algorithm  will  be  significantly  enhanced  and  properly evaluated  in  order  to  be  ready  for  real-life  deployment  and  usage.  More  precisely,  several possibilities  for  increasing  its  efficiency  will  be studied  and  the  most  suitable  will  be implemented  in  order  to  obtain  an  algorithm  that  finds  driving  strategies  in  real  time.  In addition,  human  driving  characteristics  will  be  obtained,  analyzed  and  properly  included  in MODS  to  obtain  human-like  driving  strategies.  Moreover,  the  information  about  other  vehicles on  the  route  (e.g.,  the  position  and  velocity  of  the  neighbor  vehicles) will  be  integrated  in MODS.  Furthermore,  MODS  will  be  integrated  and  tested  in  a  near-real-life  environment, consisting  of a  dedicated  hardware  platform  and  a  software  simulator,  which  is  able  to  simulate  traffic involvement, safety aspects, unexpected situations, traffic fluidity, fuel consumption, etc.

Range on year

0,75 FTE

 

Research organisations

Researchers

Citations for bibliographic records

 

Project phases and their realization

Timetable
WP1 Requirements analysis (March 2016 – June 2016)
WP2 Speedup of the MODS algorithm (July 2016 – May 2017)
WP3 Additional objectives (July 2016 – May 2017)
WP4 Driver's behavior (July 2016 – May 2017)
WP5 Integration and evaluation (July 2016 – February 2018)
WP6 Promotion (March 2016 – February 2018)

Deliverables
D1 Report on requirements analysis and definition of algorithm improvements (October 2016)
D2 Prototype of the enhanced algorithm (May 2017)
D3 Final version of the algorithm (February 2018)

 

Project bibliographic references

  • E. Dovgan, J. Sodnik, I. Bratko, and B. Filipič, Multiobjective discovery of human-like driving strategies, Proceedings of the 2017 Genetic and Evolutionary Computation Conference Companion, 8 pages, 2017.
  • K. Trontelj, T. Čegovnik, E. Dovgan, and J. Sodnik. Evaluating safe driving behavior in a driving simulator, Proceedings of the 7th International Conference on Information Society and Technology, pp. 299-302, 2017.
  • E. Dovgan, J. Sodnik, I. Bratko, and B. Filipič, Optimization of time and fuel consumption in human driving models, Proceedings of the 20th International Multiconference Information Society - IS 2017, vol. A, pp. 35-38, 2017.
  • T. Čegovnik, K. Trontelj, E. Dovgan, and J. Sodnik. Evaluation of driver's behavior and traffic safety in a driving simulator, Proceedings of the Twenty-sixth International Electrotechnical and Computer Science Conference ERK 2017, pp. 264-267, 2017.
  • E. Dovgan, Multiobjective discovery of driving strategies using the SCANeR Studio, Proceedings of the 19th International Multiconference Information Society - IS 2016, vol. A, pp. 21-24, 2016.
  • E. Dovgan, Discovering human-like driving strategies with a multiobjective optimization algorithm, JSI Technical report no. 12448, 2018.
  • E. Dovgan, MOHDS, Software, Ljubljana: Jožef Stefan Institute, 2018.

 

FINANCED BY

 

  • Collaborators