TOP: Vehicle Trajectory based Driving Speed Optimization Strategy for Travel Time Minimization and Road Congestion Avoidance

Transcription

1 TOP: Vehicle Trajectory based Driving Speed Optimization Strategy for Travel Time Minimization and Road Congestion Avoidance Authors: Li Yan and Haiying Shen Presenter: Ankur Sarker IEEE MASS Brasília, Brazil October 2016

2 Why is traffic congestion control pivotal? 2

3 Why is traffic congestion control pivotal? 3

4 Why is traffic congestion control pivotal? 4

5 Why is traffic congestion control pivotal? 5

6 6

7 7 1 Use signal to schedule passing of vehicles

8 8 1 2 Use signal to schedule passing of vehicles Use vehicle s driving info to optimize speed

9 Problem 9

10 Problem 10 Overlook the possible road congestion generation in the future

11 Problem 11 Overlook the possible road congestion generation in the future

12 Problem 12 Overlook the possible road congestion generation in the future

13 TOP: Trajectory based speed OPtimization 13

14 14 TOP: Trajectory based speed OPtimization Adjust vehicles mobility to alleviate road congestion globally

15 Overview 15 Trace analysis and supportive findings for TOP Design of TOP Experimental results Conclusion with future directions

16 16

17 17 Vehicles concurrent competition for few popular roads Excessive usage of the roads

18 18 Vehicles concurrent competition for few popular roads Excessive usage of the roads Distribute vehicle traffic evenly in all road segments Avoid road congestion and increase the utilization of road network

19 19

20 20 Vehicles temporal preference on roads High vehicle density during some times

21 21 Vehicles temporal preference on roads High vehicle density during some times Allocate the usage of roads to different time slots Avoid high vehicle density during some times (e.g., rush hours)

22 Gaming process 22

23 Future vehicle density prediction 23 Trajectory calculation For a road segment: Estimated total travel time:

24 Future vehicle density prediction 24 Trajectory calculation For a road segment: max m li / vi, 0 di di min m jam i i / i, i i i jam, di di t l v d d d Estimated total travel time: T i M i t m 1 m

25 Future vehicle density prediction 25 Trajectory calculation For a road segment: max m li / vi, 0 di di min m jam i i / i, i i i jam, di di t l v d d d Estimated total travel time: T i M i t m 1 m Travel times follow normal distribution, and are i.i.d.

26 Future vehicle density prediction 26 Road vehicle density calculation For a road segment: N si e s i 1 k i j j k 1 d P ( T t - t ) N is the number of vehicles that will pass e s i during tj, t s j

27 Future vehicle density prediction 27 Safety estimation For a road segment: p W w Tj j w 1 i e s W t j - t j ( ) which is the accident probability of s during the j th interval i

28 29 For central server: N S L( d ) = d v i 1 i i

29 30 For central server: For drivers: N S L( d ) = d v i 1 i i j j j F( v,a, p ) =U ( v,a, p ) -U ( d,v, p ) i i i s i i i r i i j 1 j = iln( vi pi ) - pi dvi (,, j if vi i pi ) i s.t. vi vi max

30 31 Driving speed optimization gaming 1. The central server offers densities: ln( 1), 1,, D d = u d u n u c 1

31 Driving speed optimization gaming The central server offers densities: ln( 1), 1,, D d = u d u n u c 1 2. For each d u, each vehicle chooses speed by: j v = arg max F( v,, p ) ku k k k k v v k k max k

32 Driving speed optimization gaming The central server offers densities: ln( 1), 1,, D d = u d u n u c 1 2. For each d u, each vehicle chooses speed by: j v = arg max F( v,, p ) ku k k k k v v k k max k 3. The central server finalizes the expected vehicle density: d = arg max L( d ) arg max d v l u u iu d D d D N u u S

33 Driving speed optimization gaming The central server offers densities: ln( 1), 1,, D d = u d u n u c 1 4. Each vehicle updates speed according to the new vehicle density 2. For each d u, each vehicle chooses speed by: j v = arg max F( v,, p ) ku k k k k v v k k max k 3. The central server finalizes the expected vehicle density: d = arg max L( d ) arg max d v l u u iu d D d D N u u S

34 Performance evaluation 35 Vehicle mobility traces

35 Performance evaluation 36 Vehicle mobility traces Rome [1]: 30-day taxi trace with 315 taxis and 4638 landmarks

36 Performance evaluation 37 Vehicle mobility traces Rome [1]: 30-day taxi trace with 315 taxis and 4638 landmarks San Francisco [2]: 30-day taxi trace with 536 taxis and 2508 landmarks [1] R. Amici, M. Bonola, L. Bracciale, P. Loreti, A. Rabuffi, and G. Bianchi, Performance assessment of an epidemic protocol in VANET using real traces, in Proc. of MoWNeT, [2] M. Pió rkowski, N. Sarafijanovic-Djukic, and M. Grossglauser, A parsimonious model of mobile partitioned networks with clustering, in Proc. of COMSNETS, 2009.

37 Performance evaluation (cont.) 38 Metrics

38 Performance evaluation (cont.) 39 Metrics Average vehicle speed Average flow rate Average driving time Average driver satisfaction

39 Performance evaluation (cont.) 40 Rome (Ave. vehicle speed + Ave. flow rate):

40 Performance evaluation (cont.) 41 Rome (Ave. vehicle speed + Ave. flow rate): TOP>RealSpeed>Signal

41 Performance evaluation (cont.) 42 Rome (Ave. vehicle speed + Ave. flow rate): TOP>RealSpeed>Signal TOP>RealSpeed>Signal

42 Performance evaluation (cont.) 43 Rome (Ave. driving time + Ave. driver satisfaction):

43 Performance evaluation (cont.) 44 Rome (Ave. driving time + Ave. driver satisfaction): Signal>RealSpeed>TOP

44 Performance evaluation (cont.) 45 Rome (Ave. driving time + Ave. driver satisfaction): Signal>RealSpeed>TOP TOP>RealSpeed>Signal

45 Conclusions 46

46 Conclusions Vehicle traffic has characteristics that can easily lead to concurrent competition of roads, namely congestion.

47 Conclusions Vehicle traffic has characteristics that can easily lead to concurrent competition of roads, namely congestion. 2. The formulated non-cooperative Stackelberg game between vehicles and a central server can evenly distribute traffic and avoid congestion.

48 Conclusions Vehicle traffic has characteristics that can easily lead to concurrent competition of roads, namely congestion. 2. The formulated non-cooperative Stackelberg game between vehicles and a central server can evenly distribute traffic and avoid congestion. 3. Majority of the vehicles have social patterns, which may be exploited to further avoid the generation of traffic congestion

49 50 Thank you! Questions & Comments? Li Yan, Ph.D. Candidate Pervasive Communication Laboratory University of Virginia