1. Motion Planning for Multiple Autonomous Vehicles
Fuzzy Logic
Rahul Kala
Presentation of paper: R. Kala, K. Warwick (2015). Reactive Planning of Autonomous Vehicles for Traffic Scenarios. Electronics 4,
April, 2013

2. Why Fuzzy Logic? Computational Time
Work with partially known environments
Issues
Completeness
Optimality

3. Key Contributions Design of a Fuzzy Inference System for the problem.
Design of a decision making module for deciding the feasibility of overtaking purely based on the vehicle distances and speeds.
Design of an evolutionary technique for optimization of such a fuzzy system.
Using the designed fuzzy system enabling vehicles to travel through a crossing by introducing a virtual barricade.

4. Fuzzy Inference System
Codify the immediate scenario to a few inputs.
Decide the actions to be taken and hence design the outputs.
Think of various scenarios and the associated inputs/outputs and generalize them as rules
Design Methodology
Formulate Inputs
Formulate Outputs
Steering
Speed
Design Rules

5. Discrete Valued/ Strategy Inputs
Continuous Valued
Angle deviation from road
Distance from left boundary/ obstacle
Distance from right boundary/ obstacle
Distance from front vehicle/ boundary/ obstacle
Side: distance of vehicle in wrong side
Discrete Valued/ Strategy Inputs
Turn to avoid obstacle/ overtake vehicle in front
Requested turn: turn to enable another vehicle to overtake

6. Minimal of twin distance inputs is taken
Some Inputs γi θi
Obstacle
Left Distances
Right Distances
Front Distances
Deviation = γi – θi
Minimal of twin distance inputs is taken

7. Turn to avoid obstacle
If front distance from left corner is less than front distance from right corner, turn right; and vice versa
Heuristic holds from most small obstacles/general scenarios

8. Overtaking
Check if the front vehicle (1) is slower and needs to be overtaken by the vehicle being planned (2)
Check if road is wide enough to safely accommodate both (1) and (2)
Check if overtaking is feasible with any other vehicle (3) on the road while (2) overtakes (1)
Initiate the overtake
(2) decides the side of overtake
(2) steers on decided side
(1) steers on opposite side
Check whether cooperation of (3) is required or not
In case yes, (3) steers depending upon the scenario

9. Feasibility criterion with 3 vehicles
Assumption: Road not wide enough for multiple (>3) vehicles to lie side-by-side
The vehicles are projected to travel straight subsequently (else would require to adjust for overtaking)
Condition 1: (1), (2) and (3) can simultaneously lie side by side along the road, OR
Condition 2: (2) can complete overtake of (1) within the time (3) does not lie in the overtaking zone

10. Cooperation
Whether (3) needs to steer in a particular direction to enable overtake?
Cooperation required in case the vehicles need to align to fit within road width (condition 1)
(3) moves opposite to the location of (1) and (2)

11. Vehicle Following
In case of infeasibility, the vehicle would follow the vehicle in front
If any oncoming vehicle/any other vehicle causing overtake infeasibility passes, overtake initiates
Vehicle can drive in the wrong side and does not consider the vehicle in front (both inputs disabled)
When (2) is ahead of (1), the inputs are enabled again

12. Crossing
Add virtual barricade as boundaries on road not used in navigation
Make the vehicle go by designed fuzzy planned – overtaking disabled
In absence of traffic lights first come first serve sequence followed

13. Crossing
Left Boundary
Right Boundary
Barricades

14. Evolution of fuzzy planner
Optimization using Genetic Algorithm
Complete design using Genetic Algorithm would be computationally expensive
Initial fuzzy planner designed by human based on sample scenarios
Altering optimization of rules and membership functions for a few cycles
Rule optimization can increment/decrement any antecedent/consequent by a unity
Membership function optimization can move any membership function parameters within a narrow region.
Fitness Function: Minimize time, minimize collisions, minimize time in wrong side, minimize safety distance breach

15. Evolution of fuzzy planner
F ← Human Designed Fuzzy Planner
While no of cycles are not met
F ← Tune Rules (F)
F ← Tune Membership Function Parameters (F)
Fitness Evaluation
Simulation
Map
Return F
End

16. Results – Single Vehicle

17. Results – Vehicle Avoidance

18. Results – Multi Vehicle

19. Results – Multi Vehicle

20. Results - Crossing

21. Analysis
Two vehicle scenarios
Single vehicle scenarios

22. Analysis
Overtaking scenarios

23. Angle Deviation from Road
Two vehicle scenarios
Single vehicle scenarios

24. Angle Deviation from Road
Overtaking scenarios

25. Thank You Acknowledgements:
Commonwealth Scholarship Commission in the United Kingdom
British Council
Thank You