Intelligent Driving Agents Intelligent Driving Agents “Microscopic traffic simulation with reactive “Microscopic traffic simulation with reactive driving agents” driving agents” ITS 2001 Conference, Oakland ITS 2001 Conference, Oakland Patrick Ehlert and Leon Rothkrantz Patrick Ehlert and Leon Rothkrantz August 28th, 2001 August 28th, 2001 Delft University of Technology
Overview of presentation Overview of presentation � Project Project � � Design of driving agent Design of driving agent � � Implementation in prototype simulator Implementation in prototype simulator � � Results, conclusions and future work Results, conclusions and future work � TU Delft 1 1
Project Project � Create intelligent agent with human Create intelligent agent with human- -like like � driving behavior driving behavior � Goals: Goals: � 1. Perform reasoning from local viewpoint 1. Perform reasoning from local viewpoint 2. Investigate interaction between drivers 2. Investigate interaction between drivers 3. Create flexible and realistic traffic simulator 3. Create flexible and realistic traffic simulator � Used Used tactical tactical- -level level reasoning reasoning � TU Delft 2 2
Design: driving agent Design: driving agent � Perform human Perform human- -like tactical driving like tactical driving � � Real Real- -time control of vehicle time control of vehicle � � Expandable and flexible functionality Expandable and flexible functionality � TU Delft 3 3
Design: driving agent (continued) Design: driving agent (continued) Vehicle Sensors Communication Controller & Arbiter Environment Supervisor / Memory other agents Behavior rules Parameters Car following Overtaking Traffic Behavior rules lights Road Collision following avoidance TU Delft 4 4
: simulator Implementation : simulator Implementation � Decided to create new prototype Decided to create new prototype � simulation program simulation program � Time Time- -oriented simulator oriented simulator � � Kinematic Kinematic 2D motion model 2D motion model � TU Delft 5 5
Implementation: simulator Implementation: simulator Simulation Environment controller 1: update Timer Vehicles Traffic lights Roads User interface 2: visual Traffic light Intersections feedback controllers Picture of environment Simulated objects TU Delft 6 6
Implementation: agent Implementation: agent Environment Agents b: send orders Reasoning c: sleep Vehicles Sensors Traffic lights Roads a: get information Traffic light Intersections controllers Simulated objects TU Delft 7 7
Implementation: agent’s rules Implementation: agent’s rules � Behavior Behavior rules are directly coded into the rules are directly coded into the � program for fast performance program for fast performance example: If If (agent speed < preferred speed) (agent speed < preferred speed) example: then Accelerate (normal) Accelerate (normal) then TU Delft 8 8
Implementation: example Implementation: example TU Delft 9 9
Results Results � Simulation prototype with Simulation prototype with � � Up to 40 vehicles (agents) in real Up to 40 vehicles (agents) in real- -time time � � Human Human- -like driving behavior like driving behavior � � Interaction between drivers Interaction between drivers � TU Delft 10 10
Conclusions Conclusions � Advantages agent Advantages agent- -based simulation based simulation � • increased realism increased realism • • allows more flexibility allows more flexibility • • distributed processing possible distributed processing possible • � Disadvantages Disadvantages � • increase computational load increase computational load • • many parameters, more difficult validation many parameters, more difficult validation • TU Delft 11 11
Future w ork Future w ork � Expand simulator and agent functionality Expand simulator and agent functionality � � Use distributed approach (more agents) Use distributed approach (more agents) � � Nanoscopic Nanoscopic simulation simulation � � Use agent model to control real vehicles Use agent model to control real vehicles � More info at More info at http://www.kbs http://www. kbs. .twi twi. .tudelft tudelft. .nl nl/People/Staff/P.A.M.Ehlert/ /People/Staff/P.A.M.Ehlert/ai ai/project_IDA.html /project_IDA.html TU Delft 12 12
Recommend
More recommend
