Densité de Trafic Emergente pour des Véhicules Intelligents Communiquants Guidés par Heuristique Density of Emergent Traffic of Heuristically-Driven Intelligent and Communicating Vehicles
Philippe Morignot Oyunchimeg Shagdar Fawzi Nashashibi APIA 2015
Rennes, July 2015
Research motivation --- Path planning of Intelligent vehicles in urban roads --• V2X communication • Research on the domain of wireless communications • Limited study about the impact of communication on the traffic behavior
• Path planning • Research on the domains of AI, robotics • Assumption of perfect communication Impact on individual’s knowledge
V2X communications
Path planning
Impact on traffic density
Research Objective Studying the collective behavior of vehicles, which individually plan their path based on V2V information exchange. APIA 2015
Rennes, July 2015
2
1
Intelligent vehicle: Finite state automaton 1.
Initial state: Randomly assigned position and goal (destination)
2.
If the vehicle is on a straight road accelerate (speed limit) or decelerate (collision avoidance, or red traffic light).
3.
If the vehicle reaches an intersection choose the exit (N/E/S/W) of the intersection based on information obtained from the V2V communication and the path-planning heuristics.
4.
If the vehicle reaches its goal location, randomly assign a new goal. Rennes, July 2015
APIA 2015
4
V2V Communications -- Channel model-• Two-ray ground ground reflection model (TRG)
V3
RayTrace (V3)
-40 -50 -60 -70 -80 -90 -100 -110 -120 -130 -140
Signal strength [dBm]
V2
d V1
d Lfs= 20 log10 4π , d ≤ d λ Ltrg[dB] = 2 Lfs= 20 log 4π d , otherwise 10 ht hr
RayTrace (V2)
TRG
Receive threshold (6 Mbps)
0
10
20 30 40 50 60 70 80 d (V1's distance from the junction) [m]
90
• TRG can be used for signal estimation for vehicles on the same road • It is difficult to expect communications between vehicles on crossing roads
APIA 2015
Rennes, July 2015
6
2
V2V Communications – MAC Model -• IEEE 802.11p MAC: same as IEEE 802.11e • Prioritized channel access for different access categories
• Transmission of CAMs • A single AC • Broadcast: no retransmission 1-D Markov-Chain
1/W 1/W
1
1/W 0
1/W 1
1/W 2
1-pk
1-pk
pk
pk
Channel access probability (saturated channel)
τ s = b(0) = 1+
W-2 1-pk
1-pk
pk
Channel access probability
−1
W −1 2(1− pk )
τ = (1− exp (−λYs )) × τ s
1-pk
W-1 pk
Comm. success probability
Ps = (1− τ )
N−1
Rennes, July 2015
APIA 2015
7
Path-planning heuristics • Compass: Take the shortest path to the goal without considering vehicle density • Ant: Take the more popular road, which leads to the goal. • No-ant: Take the less congested road, which leads to the goal.
a1 < a2
APIA 2015
Goal
p1
a1
a2
p2
Compass : a1 (p1 is always chosen) Ant : max(#Cars(a1), #Cars(a2)) (p1 or p2) No-Ant : min(#Cars(a1), #Cars(a2)) (p1 or p2)
Rennes, July 2015
8
3
Experiments: NetLogo multi-agent simulator • Scenario: Manhattan streets, traffic lights, traffic regulations, up to 400 vehicles • Communications: • No communications. • Ideal communications: no communications error • Realistic communications: message delivery probability for a given transmitted CAM is calculated based on propagation/MAC models
• Path planning • Compass: No communications • Ant: with/without communication • No-Ant: with/without communications Rennes, July 2015
APIA 2015
9
120 Ant (Ideal-comm) 100
No Ant (Ideal-comm)
80
Ant (Real-comm) No Ant (Real-comm)
60
No comm 40 20 0 50
100
150 200 250 300 350 Total number of vehicles
400
Probability of successful recep on
STD of the nubmer of vehicles on individual streets
Communications impact 1 Ant (Real-comm) 0.9
No Ant (Real-comm)
0.8 0.7 0.6 0.5 50
100
150 200 250 300 Total number of vehicles
350
400
Conclusion • Ant/No-ant strategies show their expected behavior when comm. is ideal • And/no-ant strategies fail to show their expected behavior when comm. is realistic. • Communications failure APIA 2015
Strategy failure Rennes, July 2015
10
4
Emergent behaviors Ant (Ideal-comm)
No Ant (Ideal-comm)
Ant (Real-comm)
No Ant (Real-comm)
M ean speed over the last 6000 ticks
No comm 12 11 10 9 8 7 6 5 4 50
100
150 200 250 300 Total number of vehicles
350
400
Conclusion • Communications has a great impact on the collective behavior • Ideal communications: No-ant behavior provides faster velocity (better to have an information) • Realistic communications: the difference between ant and no-ant ant behaviors is small APIA 2015
Rennes, July 2015
11
Rennes, July 2015
13
Thank you!
Oyunchimeg SHAGDAR
[email protected]
APIA 2015
5
Path-planning for Intelligent Vehicles • Grid-based with a search algorithm (e.g., A*). • The finer the grid, the longer the search.
• Potential fields : attracted to goal, repulsed from obstacles. • Might be trapped in local minima.
• Paving-based search. Generate 2 sub-pavings such that X- ⊂ Cfree ⊂ X+ • Sampling algorithms : sample N configurations and retain those in Cfree. If there is a path between 2 samples, retain it ; otherwise, resample. • Works well for high dimensional problems.
APIA 2015
Rennes, July 2015
14
6