Automatic Merge Control Algorithms Ashish Gudhe Roll. No. 05305028 Guide :- Prof. K. Ramamritham.

Automatic Merge Control Algorithms

Ashish Gudhe

Roll. No. 05305028

Guide :-

Prof. K.

Ramamritham

Roadmap

Aim of this project

Introduction to Automatic Merge Control(AMC)

Existing AMC algorithms

Proposed AMC algorithms

Experiments

Conclusion

Future Work

Video

Reference

Aim of this project

Understand AMC

Primary Aim : Safety at intersection of lanes Secondary Aim :

Increase Traffic Throughput Minimize Time To Merge

Design AMC algorithms. Study performance of the algorithms. Perform experiments on vehicular platforms.

Introduction to AMC system

What is Automatic Merge Control (AMC)?

Automotive Application for automated merging of

vehicles in the intersection region.

Determines merge order of vehicles.

Safety-Critical Application.

Some definitions in AMC system

Time To Merge (TTM):- It is the time required by a

vehicle to reach the intersection region from current position.

Merging Sequence:- It is the order in which vehicles

merge at the intersection.

Area of Interest (AoI):- The area defined by radius R

where all vehicles in this area become part of AMC algorithm.

Safety Distance :- The minimum separation the

vehicles need to maintain at any instant of time.

AMC Algorithms

Existing algorithms

Optimization formulation

Head of Lane Approach (HoL1)

Virtual Vehicle based merging.

Proposed algorithms

Virtual Vehicle based merging.

Head of Lane Approach (HoL2).

All Sequences Minimal Cost (ASMC).

Optimization Formulation[6]

Merging problem is formulated as an optimization problem.

Input : Vehicles’ profiles.

Output: TTM of each vehicle.

Objective function is to minimize the average TTM.

Constraints:-

Precedence Constraint

Mutual Exclusion Constraint

Lower bound on TTM

Drawbacks :

Non-linear constraints. Global optimum not guaranteed.

Computationally intensive.

HoL Approach[7] : HOL1

• Head of lane is the leader vehicle.

• Merge sequence is generated iteratively by inserting selected head vehicle.

• Merging Decision : To select either V11 or V21.– Accelerate both the vehicles to tolerable limits.

– Insert Nearest Head vehicle in case of interference. Choose the next vehicle in the lane as new head of lane.

– Insert the vehicle with smaller TTM and the next vehicle in same lane becomes new head of lane.

Virtual Vehicle based Merging

• Virtual vehicle[4] is an image of actual vehicle mapped on other lane.

• Longitudinal control using Adaptive Cruise Control (ACC)[1].

• The virtual vehicle becomes lead vehicle.

• ACC with virtual vehicle.

• Selection of vehicle as a virtual vehicle depends upon two main criteria :-

• Spatial proximity• Temporal proximity

Virtual Vehicle based Merging

Choice of virtual vehicle

depends on following criteria:-

Spatial Proximity e.g., V2 is mapped as V2’

Temporal Proximity e.g., V1 is mapped as V1’

Virtual vehicle based merging

• Zonal distribution of AoI• Zone 3 monitors the number of

vehicles entering the AoI.• Zones 1 and 2 define the two modes of

system operations (communication, vehicle mapping, ACC, etc.)

• The frequency of operations of vehicles in Zone 1 can be double that of in

Zone 2. • Area based mode change.• Better control expected close to

intersection region. • Zone sizes can be made flexible to

handle traffic of various natures.

All Sequences Minimal Cost (ASMC) Approach

Generates all possible valid merge sequences. Outputs the best merge sequence i.e. one with

minimal merging cost. Recursively computes the valid sequences by

dividing the merging problem with n vehicles to merging problem with (n-1) vehicles.

Guarantees optimal solution. Computationally intensive and space

consuming. Benchmark for comparing performance of other

approaches.

ASMC algorithm short description

where:-• S1 = set of vehicles on lane 1• S2 = set of vehicles on lane 2• refVehicle : reference vehicle w.r.t which behavior of remaining

vehicles is computed.

HoL2 approach

Cascading effect

considered

Merge sequence

generated based upon

effect on the subsequent

vehicles.

Effect in terms of TTM

Effect in terms of deceleration

Effect in terms of number of

vehicles being affected

Extensions to AMC algorithms : Handling continuous streams of

vehicles• AMC algorithms take static snapshot of vehicles

• Merge sequence generated considering this snapshot

• How to handle continuous stream of vehicles entering Area of Interest?– Time based approach where snapshot is taken in regular

intervals of time.– Zonal distribution of Area of Interest where the AoI is

divided into zones and snapshot timing depends on certain criteria.

Zonal distribution of AoI

Z1: Zone 1 closest to

intersection region. Vehicles’ profile remain unchanged in this zone

Z2: Zone 2 of which snapshot

is taken. All vehicles in this zone are part of AMC algorithm.

Z3: Covers entire AoI along

with Z1 and Z2. Farthest zone

which tracks new vehicles that

are about to enter Z2.

Extensions to AMC algorithms cont…

Snapshot timing

After a new vehicle enters zone

Z2

more computations

high prob. of same vehicle

being included in

successive snapshot.

After zone Z2 of any lane

becomes empty :

lesser computations

low prob. of same vehicle

being included in

successive snapshot.

Virtual Vehicle experiment

ACC enabled vehicle VACC

Cruise controlled vehicle VCC

Local learning with position feedback

Initial distance from intersection is fixed :

SACC=1000mm and SCC=500mm.

Current position computed from position

feedback.

Experimental Robotic Vehicles

Virtual Vehicle experiment cont…

Spatial proximity is used as the criteria for

virtual vehicle mapping. Hence, VCC is mapped

as virtual vehicle ahead of VACC.

VCC communicates its current location to VACC

VACC computes the separation distance from the

virtual vehicle.

VACC performs ACC with the virtual vehicle.

Virtual vehicle merging results

Velocity(VACC)=100 mm/sec and Velocity(VCC)=50mm/sec and desired time gap was set to 1sec.

Final velocity of VACC=50mm/sec approx.

The vehicle VACC follows the virtual lead vehicle with time gap between 1 to 1.5 sec.

Virtual vehicle merging results

The initial distances of VACC and VCC are 1000mm and 500mm

respectively from intersection region.

Y-axis denotes the distance from intersection region.

Safety is ensured at the intersection region.

C++ Simulation experiments (case 1)

Vehicle parameters :

Velocity bounds = [0,27] m/s

Acceleration bounds = [-4,4] m/s2

Safety distance = 5m.

Vehicle profiles at time t=0

Comparative results (case 1)

• Total TTM– ASMC = 14.12 sec– HoL1 = 15.011sec– HoL2 = 14.12 sec

• Here HoL2 performs better than HoL1.

Graphs (case 1)

ASMC HoL2

C++ Simulation experiments (case 2)

Vehicle parameters :-Velocity bounds = [0,27] m/s

Acceleration bounds = [-4,4] m/s2

Safety distance = 5m

Vehicle profiles at time t=0

Comparative results (case 2)

• Total TTM– ASMC = 17.63 sec– HoL1 = 17.84 sec – HoL2 = 18.65 sec

• Here HoL1 performs better than HoL2

Graphs (case 2)

ASMC HoL1

Conclusions

Comparative study of Automatic merge

control algorithms.

Few approaches are proposed which

ensure safety and high traffic

throughput.

The HoL1 approach is observed to

perform better than HoL2 in certain

scenario and vice-versa.

VIDEO

Future work

To study the performance of the AMC algorithms on continuous streams of vehicles.

To perform experiments on robotic vehicular platforms.

To design and implement a decentralized controller for decision making i.e. to allow vehicles to take decision.

To study vehicle-to-vehicle communication aspects.

Post-merging safety and stability.

References[1] Gurulingesh G., Neera Sharma, K. Ramamritham and Sachitanand M. Efficient

Real-Time Support for Automotive Application : A Case Study. In Proceedings of the

RTCSA 2006, Sydney, Australia, Aug 2006.

[2] Xiao-Yun Lu. and Hedrick K.J. Longitudinal control algorithm for automated

vehicle merging. In Proceedings of IEEE Conference on Decision and Control,

volume 1, pages 450-455,2000.

[3] Hossein Jula, Elias B. Kosmatopoulos, and Petros A. Ioannou. Collision

Avoidance Analysis for Lane Changing and Merging. In IEEE Transactions on

Vehicular Technology, volume 49, pages 2295-2308, Nov 2000.

[4] A. Uno, T. Sakaguchi, and S. Tsugawa. A merging control algorithm based on

inter-vehicle communication. In Proceedings of IEEE/IEEJ/JSAI International

Conference on Intelligent Transportation Systems 1999, pages 783-787, 1999.

[5] Steven E. Shladover Xiao-Yun Lu., Han-Shue Tan. and J. Karl Hedrick.

Implementation of longitudinal control algorithm for vehicle merging. In

Proceedings of AVEC 2000 5th International Symposium on Advanced Vehicle

Control, Ann Arbor, Michigan, Aug 2000.

[6] Gurulingesh G., J Bharadia and K. Ramamritham. Towards Intelligent Vehicles :

Automatic

Merge Control. In RTSS Workshop 2006, Brazil, Dec 2006.

[7] Vipul Shingde. Report on Automatic Merge Control, Indian Institute of

Technology, Bom-

bay. 2006.

Thank you

Prof. A A Diwan

Prof. Kavi Arya

Prof. Krithi Ramamritham

Gurulingesh R

Vipul Shingde

Sachitanand Malewar

Components of AMC System

Local Learning (e.g. GPS, roadside sensors, etc.)

Communication capability

Inter-Vehicle communication

Vehicle-Roadside communication

Sensors

Speed

accelerometer

Vehicle Profile

Controller Centralized (Intersection

manager)

Decentralized

Assumptions in AMC algorithms

Vehicles are treated as points on the lane.

Vehicles know their location from intersection region.

Intersection manager exists at the intersection region. It does all the computations and controls the merging.

Communication exists between vehicles and intersection manager.

Vehicles are initially separated by safety distance.

Vehicles’ control system is reliable i.e. vehicles follow the behavior as commanded by intersection manager.

Optimization formulation[6]

• Minimize • Subject to

– Precedence constraint where is TTM of jth vehicle of ith lane and

– Mutual exclusion constraint• Where S=safety distance

– Lower bound on TTM• is max. velocity

1 2

1 21 11 2

1( )*( )

i

m m

m mji j

f t tm m

( 1)ij i jt t ijt1 1ij m 1 2i

1 21 2

, ( )min( , )k l

k l

Sk l t t

v v

, ijMAX

Sj t

V 1 2i

MAXV