1. Behavior Control of Virtual Vehicle
Hongling Wang
April 21, 2003

2. Introduction Purpose of behavior control Behavior control is complex
Run a virtual vehicle on a road network
Following traffic rules
A vehicle should be able to get anywhere in the road network
Behavior control is complex
Divided into basic component behaviors
Integrate all the basic components

3. Components of Vehicle Behavior
Cruising behavior
Vehicle drives at desired speed
Following behavior
Vehicle keeps a safe distance behind its leader
Intersection behavior
Vehicle traverses intersections safely
Obeys traffic signals
Respects right of way
Lane changing behavior
Vehicle leaves the current lane and enters an adjacent target lane

4. Behavior and Kinematics
Behavior sets control parameters
Acceleration
Driving curvature
Kinematics moves a vehicle to a new position according to parameter values

5. Path Path, a ribbon composed of road lanes and intersection corridors
Path used to guide vehicle moving
Path forms a consistent frame of reference
Pursuit point on path centerline
Path is an interface between a vehicle and outside world

6. Path (Cont.) Path simplifies behavior control
Driving curvature determined by path
Acceleration determined by behaviors
Path provides a basis for spatial relationship

7. Cruising behavior Determines desired speed
Compare current speed with desired speed
Current speed is higher, negative acceleration
Current speed is lower, positive acceleration
Proportional controller
Reactive behavior
Decision depends only on the state at this moment

8. Following behavior Query the leader on the path of a vehicle
Compute relative distance and relative speed
Proportional-derivative controller
Contribute the acceleration if negative, discard it if positive
Reactive behavior

9. PD controller response to a slow leader
acceleration
Before critical point, positive
After critical point, negative and increasing
After negative maximum, negative and decreasing to approach 0
phases of vehicle actions
No response
Slow down to leader’s speed
Keep a safe distance from its leader

10. Integration of cruising and following behaviors
If following acceleration >0, choose cruising acceleration
If following acceleration <=0, choose smaller value among the two
Integrated behavior: a vehicle always tries to drive at a desired speed, while keeps from running into or too close to its leader

11. Intersection behavior
What a vehicle does before entering an intersection
Stop
Keep going
Stop and go alternatively
Actions chosen according to ambient traffic and traffic control signals
Sequential behavior
Decision depends on both the state in last moment and the state in this moment

12. Intersection behavior (Cont.)
An intersection is a resource
A vehicle should not enter it if it can’t leave it soon
Three sub behaviors because of different right-of-way rules
Going straight
Turning left
Turning right

13. Intersection behavior (Cont.)
Main problems
Stop a vehicle on desired position
Using state machines to control action flow
Gap acceptance
Immediate gap (e.g., turning right on RED)
Predicted gap (e.g., turning left on GREEN)

14. Stopping behavior Requirements Acceleration computation method
Inform a vehicle it is the time to decelerate
Stop a vehicle in desired position if computed acceleration applied
Keep a vehicle stopped after it stopped
Acceleration computation method
PD controller
Invariant acceleration controller

15. PD controller for stopping
Acceleration formula
Phases of vehicle actions
No response
Slow down and stop at desired position
Stay stopped at desired position

16. PD controller for stopping
disadvantages
No fully stopping, speed infinitely approaches 0
Acceleration value may be too big, if critical point is missed

17. Invariant acceleration controller for stopping
Advantages
Be able to give a full stop at desired position
Gives a reasonable acceleration in some cases where PD controller gives a too big acceleration
Disadvantage
Sensitive to small errors of both speed and distance
Conclusion: a better choice than PD controller

18. State machines for Intersection behavior
Basic states of state machines for intersection behavior
START, no response to state of control signal
CONTINUE, keep going while stopping still possible
SLOWDOWN, decelerate for stopping
STOPPED, speed is 0
END, stopping becomes impossible or is no longer necessary
One state machine built for each sub behavior

19. Gap acceptance computation
Gap is a time period within which my required space is free
A resource
Relationship between time and space
Immediate gap
Estimate when others will get to my required space
Check if it is within the gap
Predicted gap
Estimate when I will get to and leave my required space
Check if they overlay

20. Intersection behavior by simple right of way rules
Problems:
Deadlock
Starvation
Solutions
Deadlock breaking rule
Starvation avoidance rule

21. Integration of cruising, following and intersection behaviors
Before intersection behavior is activated, choose the former acceleration
After it is activated
In SLOWDOWN or STOPPED phase, choose the smaller value among the former acceleration and intersection acceleration
In other phases, choose the former acceleration
Integrated behavior: A vehicle tries to drive at a desired speed, keeps a safe distance with its leader and responds to traffic control signals on intersections

22. Lane change behavior Modeled as a sequence of four steps
Consider a lane change
Choice of a target lane
Gap acceptance
Move over to the target lane
Classified as MLC and DLC
MLC, mandatory lane change
DLC, discretionary lane change
Sequential behavior
State machine with 4 states corresponding to the 4 steps

23. Discretionary Lane Change
Consider DLC when the speed is below a desired speed
Change to a neighboring lane for opportunity to increase speed
A gap is acceptable when both lead and lag gaps on target lane are acceptable

24. Trajectories of a vehicle and its pursuit point during lane changing
Move pursuit point from center of current lane to center of target lane
Use PD controller to control lateral moving of pursuit point
Vehicle overshoots the target offset

25. Gap acceptance for lane change
Both lead gap and lag gap are acceptable
My current leader and follower are not changing to my target lane
No vehicle on another adjacent lane of my target lane is changing to my target lane

26. Integrate lane change behavior with following behavior
The concept of following leader changed
The ahead vehicle in my current lane
The ahead vehicle in my target lane if I am in lane change
The ahead vehicle whose target lane is my current lane and who is in lane change
Problem: too conservative
Solution: visibility computation

27. Visibility computation
The ahead vehicle in my current lane may be out of my way when I am in lane change
Lane change will complete sooner with visibility computation, especial when ahead vehicle is very slow

28. Take MLC into consideration
MLC is necessary
The concept and structure of path don’t support MLC efficiently
Route, a higher level conceptual structure, is necessary for MLC
Route of a vehicle is composed of roads
Relation between route and path
Route is a long term plan
Path is a short term plane
Path is built to follow route

29. Take MLC into consideration
A multiple-lane MLC is treated as multiple single-lane changes
When still far from road end, consider only DLC, not MLC
DLC consistent with target of MLC is given priority
DLC against target of MLC is given some penalty for resource requirement

30. General Behavior Integration
Acceleration combined contribution from
Cruising behavior
Following behavior
Intersection behavior
Driving curvature combined contribution from
Path following
Lane changing behavior