1. Evolving a rule system controller for automatic driving in a car racing competition Diego Perez, Yago Saez Member, IEEE, Gustavo Recio, Pedro Isasi Presented by Evgenia Dubrovsky

2. Overview  Introduction  Objectives  Controller Design  Results  Future work

3. Introduction   In 1970, the visionary Robert E. Fenton predicted how automatic vehicle guidance would evolve.   The system would be implemented in three stages.   The installation and use of various driver aids so that the driver would be a more effective decision maker and improve the performance of the driver- vehicle system.   Involve the gradual introduction of various subsystems for partial automatic control.   The transition to complete automatic vehicle control.

4. Introduction   Today, we are in the second stage:   We can buy cars with   electronic aided brake systems (ABS)   adaptive cruise controls which maintain a set distance from the car ahead, automatically accelerating or decelerating, and even applying the brakes.   All these new technical advances are a good starting point for approximating the third stage: to complete automatic vehicle control.

5. Introduction   Annual competition is organized by the DARPA for driverless cars which evaluates all the improvements.   However, researchers have a long way to go until these vehicles capable of driving in a fully automated way are actually available.   Most of the works were engineered in real prototypes, this involves two main problems:   The costs of buying and modifying a car.   The time needed for each test.   To overcome these constraints use car simulators.

6. Introduction   A. Evolutionary computation techniques applied to the automated driving.   Research how these techniques can help in automatic driving.   Some work was collected related to evolutionary computation applied to automated driving.   These works fall into two different categories:   Autonomous driving.   Vehicle features optimization.

7. Introduction A.1. Autonomous driving   This work uses modules which combine high level task goals with low-level sensor constraints.   Those modules are dependent on a large number of parameters.   PBIL is proposed for automatically setting each module’s parameters.   The evaluation function takes into account different aspects, such as serious crashes, collisions, etc.

8. Introduction A.1. Autonomous driving   For the simulation, the system uses a program called SHIVA which reproduces a microsimulation of vehicles moving and interacting in a roadway.   The algorithm can influence the vehicles’ motion sending simulated commands:   Steering   Accelerate   Brake

9. Introduction A.1. Autonomous driving   PBIL was compared to GA in the same framework:   At the end of experiments the vehicles automatically entered the test track, completed one lap and a half and finally took the exit.   Although they were not capable of avoiding collisions with other vehicles.   These experiments showed the potential for intelligence behavior in the tactical driving.

10. Introduction A.1. Autonomous driving  An approach was proposed for automated evolutionary design of driving agents.  GAs can help in designing an agent able to remotely operate a racing car.    The agent perceives the environment from a camera (position, velocity, approach angle, distance, etc).  The agent sends commands to the remote control car (forward, reverse, left, straight or right).  The analysis established that on long runs the agent’s operated car was 5% slower than the human operated one.

11. Introduction A.2. Vehicle parameter optimization   The study of the suspension system has been an interesting topic for researchers.   Because it contributes to the car’s handling and braking for good active safety and driving pleasure.   Also keeps vehicle occupants comfortable and reasonably well isolated from road noise, bumps, and vibrations.

12. Introduction A.2. Vehicle parameter optimization   A recent work showed how to optimize 66 variables from a Formula one racecar with a GA.   The parameters affect to:   Suspension system   Engine revolutions limits and gear ratios   Aerodynamics   Brake modelling   These variables were tested in Formula One ’99-’02.   A racing simulator released in 2003 with an advanced real time physics engine.

13. Overview  Introduction  Objectives  Controller Design  Results  Future work

14. Objectives A. TORCS Simulator   TORCS allows the competitors more flexibility to develop the controllers.   Advantages:   Has a high level of realism in graphics and physics.   Provides a large quantity of vehicles, tracks and controllers.   Easy to develop a controller and to integrate it within the simulator.   Disadvantage:   Memory leak each time the race is restarted.

15. Objectives B.Controllers, Sensors and Effectors  Total: 17 sensors and 5 effectors.  The sensors included information such as the angle of the car, the position, speed, location of opponents, etc.  The effectors used to drive the vehicle are the steering wheel value, both pedals (throttle and brake) and gearing change.  Five controllers were presented to the competition and three more were provided by the organizers.

16. Objectives C. Controller evaluation  The evaluation of the controllers results from their performance on three different test circuits which are previously unknown by the competitors.  The tournament is divided into two stages: 1.Every controller is thrown alone in each circuit. 2.All controllers are tested together measuring the arrival order to evaluate the drivers.  In both cases, the points assigned to each controller follows the Formula One pointscoring system (10 points for the first, 8 for the second, 6 till 1).

17. Overview  Introduction  Objectives  Controller Design  Results  Future work

18. Controller Design A. Input Data   Angle   The angle between the car direction and the direction of the track axis.   -π to π.   Discretization to range: [0, 4]   Where 0 means small angles.

19. Controller Design A. Input Data   Track Position   The distance between the car and the track axis.   -1 to 1.   Discretization to range: [0, 1]   Where 0 means centered in the track.

20. Controller Design A. Input Data  An important feature of the design is the usage of symmetry for the first two sensors.  A value of ‘−a’ and ‘a’ from any of those sensors will mean the same discretized value.  The objective of this approach:  To reduce the search space.  To avoid the algorithm learning how to face similar situations twice.

21. Controller Design A. Input Data   Speed   The speed of the car.   0 to … (km/h).   Discretization to range: [0, 3]   Where 0 means lower speed that higher values.

22. Controller Design A. Input Data  Track  Discretization to range: [0, 2]  0 means that the track edge has been detected between 20 and 100 meters.  1 means that the track edge is up to 20 meters.  2 means that no track edge is seen.

23. Controller Design B. Effectors   Throttle and brake   Discretization to range: [0, 1]   Both pedals have been codified in a common output, in order to avoid non-sense values.   For instance, a full gas value in both pedals at the same time would produce uncertain (and useless) outcomes.

24. Controller Design B. Effectors  Steering  Discretization to range: [-1, 1]  Gear  Gear changes did not take part in the learning process.  Is defined as follows:  x is the revolutions per minute of the engine.  If x > 6,000, then gear value is increased by one.  Else if x < 3,000, then current gear is decreased.  Otherwise, the gear does not change.

25. Controller Design C. Rules   The discretization applied over the input data allows to create a set of 120 rules.   The conditional part is composed of the above four sensors and the actions (acceleration, braking and steering).   This set of initial rules are the basis of the base individual.

26. Controller Design D. Base individual   Find the rule set that allows the vehicle to end the lap, centered on the track.   Minimize angle to track axis.   Symmetry: Could have been the problem?

27. Controller Design E. Evolutionary Rule System   Rule GA Individual   Rules from base individual compose the GA individual.   N rules depend on the track of base individual (N is between 10 and 20).   Include sensors and effectors.

28. Controller Design E. Evolutionary Rule System   Rule GA Step   Create a new rule by genetic operators.   Substitute the new one for the most similar in the individual.   If fitness decreases the rule stays. If not, recover the previous one.   Linear combination of lap time (40%) and damage (60%).

29. Overview  Introduction  Objectives  Controller Design  Results  Future work

30. Results  The controller developed for the competition (called DIEGO) was tested in three circuits and it obtained acceptable results in the first two.

31. Results   The usage of symmetry produced a side effect: the car drives in a smooth zig-zag trajectory centered on the track.   Because a small steering value can center the vehicle on the track, but not necessarily drive it parallel to the track axis.   The controller behaved in this way, only on specific circuits: the oval ones.   The circuits have banked bends which make zig-zag movement completely uncontrollable.

32. Results  This problem affected dramatically to the behavior of this driver in the third circuit of the test bed, as can be seen in the results.

33. Results  In the second stage, the races performed again over these circuits gave the last set of points assigned to the drivers.  These results are very similar to the results obtained in first two circuits on stage 1.  However, the results from third circuit on stage 1 were bad, so finishing in higher positions became impossible.

34. Overview  Introduction  Objectives  Controller Design  Results  Future work

35. Future work   The main objective now is to improve this controller in order to obtain better results for this new call.

36. Future work  Discretization  Use some methodology that will help us to determine the target ranges.  Symmetry  Next controller will introduce a system without symmetry, or a limited symmetry, to avoid the zig- zag movement of the vehicle.

37. Future work  Evaluation in more than one track  TORCS simulator does not easily allow restarting the race in a circuit different than the one used for starting the execution.  Use more than one circuit in the learning process collecting more efficient rules for the controller.  Opponent analysis  The vehicles hit them in the back without evading them.  Obtain a competitive advantage in the race phase.

38. The End Any Questions ?

39. The End Thank you ;)