Upload
andre-paraense
View
351
Download
1
Embed Size (px)
Citation preview
Copyright@2011 DCA-FEEC-UNICAMP
Andre Luis Ogando ParaenseDCA-FEEC-UNICAMP
Klaus Raizer, Ricardo Ribeiro GudwinDCA-FEEC-UNICAMP
BICA 2015 - Lyon,FR
A Machine ConsciousnessApproach to Urban Traffic SignalControl
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Contribution
The main contribution of this work is the application of amachine consciousness approach, based on the GlobalWorkspace Theory (Baars, 1988), to urban traffic signalcontrol, with the design and implementation of a solution tothe problem.
2 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Traffic Jam Problems
Traffic Jam Problems
3 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
CoreArchitecture
Core
proc()In
B
Out
A
Codelet
Coderack Raw Memory
Coalition
T I
Memory Object (Sign)4 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
CoreArchitecture
Architecture
Procedural Memory
Working Memory
Perceptual Memory
Sensory Memory
Motor Memory
Episodic Memory Semantic Memory
Phonologic Loop
Visual Sketchpad
Episodic Buffer
Sensory Codelets
Perceptual Codelets
Attention Codelets
Coderack Raw Memory
Emotional Codelets Learning Codelets
Consciousness
Language Codelets Consciousness Codelets
Imagination and Planning Codelets
Behavioral Codelets
MotorCodelets
5 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Purely Reactive Behavior
6 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Most Critical Traffic Situation
7 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Broadcast, interference, deliberative behavior
8 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Network Model
9 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
CST ControllerWorking Memory
Senso
ry
Cod
ele
ts
lane a: LaneSensor
In
B
Out
A
lane b: LaneSensor
In
B
Out
A
lane c: LaneSensor
In
B
Out
A
lane d: LaneSensor
In
B
Out
A
lane e: LaneSensor
In
B
Out
A
lane f: LaneSensor
In
B
Out
A
lane g: LaneSensor
In
B
Out
A
lane h: LaneSensor
In
B
Out
A
lane i: LaneSensor
In
B
Out
A
lane j: LaneSensor
In
B
Out
A
lane k: LaneSensor
In
B
Out
A
Sensory Memory
lane a : VelocitiesMO
lane a: VehicleNumberMO
lane a: DistancesFromLightMO
lane k : VelocitiesMO
lane k: VehicleNumberMO
lane k: DistancesFromLightMO
.
.
.Motor Memory
junction west: TrafficLightActuator
In
B
Out
A
junction east: TrafficLightActuator
In
B
Out
A
Moto
r C
odele
ts
junction east: PhaseMO
junction west: PhaseMO
Behavio
ral
Codele
tsC
onscio
us n
ess
Codele
tsjunction east:
OpenSlowNearLightVehiclesLanes
In
B
Out
A
junction west: OpenSlowNearLight
VehiclesLanes
In
B
Out
A
consciousness: SpotlightBroadcast
Controller
In
B
Out
A
junction east: ActivationMO
junction west: ActivationMO
10 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Simple T Network Model
11 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Simple T Result
00:00 02:00 04:00 06:00 08:00 10:00 12:00time (h)
0
20
40
60
80
100
120
140
160
mea
n tr
avel
tim
e [s
]
Fixed TimesParallel ReactiveArtificial Consciousness
12 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Corridor Network Model
13 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Corridor Result
00:00 02:00 04:00 06:00 08:00 10:00 12:00time (h)
0
60
120
180
240
300
360
420
480
540
600
mea
n tr
avel
tim
e [s
]
Fixed TimesParallel ReactiveArtificial Consciousness
14 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Manhattan Network Model
15 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Simple TCorridorManhattan
Manhattan Result
00:00 10:00 20:00 30:00 40:00 50:00 60:00time (h)
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
mea
n tr
avel
tim
e [s
]
Fixed TimesParallel ReactiveArtificial Consciousness
16 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Conclusion
A consistent gain in performance with the ”ArtificialConsciousness” traffic signal controller during all simulationtime, throughout different simulated scenarios, could beobserved, ranging from around 13.8% to more than 21%.
This work supports the hypothesis that an artificialconsciousness mechanism, using global workspace theory canbring advantages to the global task performed by a society ofparallel agents working together for a common goal.
17 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Next steps - Downtown Campinas
18 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Thank you!
email: [email protected]
twitter: @AndreLOParaense
You can find more at https://github.com/CST-Group/cst
19 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Reactive Controller Action Selection
1 Calculates its level of activation, which is given by equation 1.
a(t) =
∑c∈C
(1− αVc(t)− βXc(t))
|C |(1)
2 Determines the best phase among the possible ones.
3 Goes back to 1.
20 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Reactive Controller Action Selection
1. Lane activation 2. Junction East 3. Best phase
∑c∈C
(1− αVc(t)− βXc(t))
(2)
∑i=1→n
AT (i)
n(3)
∑tl∈G
AT (tl) (4)
AT(g) = 0.09AT(h) = 0.3AT(i) = 0.85AT(j) = 0.05AT(k) = 0.13
ATJe = 0.858 Possible phases1. G,G,R,G,G = 0.572. G,R,G,R,R = 0.943. R,R,G,R,G =0.98Best phase
21 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Machine Consciousness Controller Behavior
1 Defines the junction codelet with greater activation level,which gains access to conscious global workspace whilerespecting a minimum threshold. If none of the codeletsreaches the threshold, the system works unconsciously andglobal workspace remains empty.
2 Broadcasts the sensory information of the conscious codelet.
3 Goes back to 1.
22 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Broadcast interference rules
1 If one incoming lane of my junction is topologically connectedto one incoming lane of the conscious junction that has a redlight in its chosen phase, I must close it with a red light.
2 If one incoming lane of my junction is topologically connectedto one incoming lane of the conscious junction that has agreen light in its chosen phase, or if it is not connected at all,I must open it with a green light.
23 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Machine Consciousness Definition
Spotlight
24 / 25
ContributionMotivation
Cognitive Systems ToolkitMachine Consciouness Controller
ResultsConclusion
Q&A
Coalitions in CST
proc()In
B
Out
A
Codelet
Coderack Raw Memory
Coalition
T I
Memory Object (Sign)25 / 25