Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Mul$agent Simula$on on StarBED Deploying Agent-‐based Traffic Simula$on on StarBED
Rafik HADFI <[email protected]> Nagoya Ins$tute of Technology, Japan
October 20th, 2015
Summary 1. Mo$va$ons 2. Architecture 3. The “agent” paradigm 4. MAS on StarBED stack 5. Traffic simula$on
Mo$va$ons
• Agent-‐based, Large-‐scale, Distributed Simula$on of Complex Systems
• Case: traffic simula$on • Simula$on of complex behaviors using an agent-‐based model.
These behaviors include: – Mobility – Driving – Traffic synchroniza$on – Sensor networks
Mo$va$ons
• A generic framework for programmable agent-‐based components – Mul$layered, with different levels of abstrac$ons referring to different simula$on layers
– Reusable API for both distributed and non-‐distributed simula$ons
– Genera$on of realis$c simula$on data
• A testbed for general purpose Computa$onal Intelligence – coordina$on – collabora$on – automated nego$a$on – etc.
Architecture
• Mul$layered architecture built on a map
Geographical Layer OSM
Architecture
Geographical Layer OSM
Architecture
Architecture
Traffic Synch Layer Traffic light
Geographical Layer OSM
Architecture
Traffic Synch Layer Traffic light
Sensor Layer Sensor network
Geographical Layer OSM
Architecture
Traffic Synch Layer Traffic light
Sensor Layer Sensor network
Agent Layer Driver agent
Mobile app agent
Geographical Layer OSM
Architecture
Traffic Synch Layer Traffic light
Sensor Layer Sensor network
Agent Layer Driver agent
Mobile app agent
Consensus Layer Rou$ng
Nego$a$on Coopera$on Collabora$on
Geographical Layer OSM
• Reac$ve, autonomous, collabora$ve, and goal-‐oriented agents – autonomous: parallel and distributed deployment (StarBED)
• Microscopic (autonomous) and macroscopic (collec$ve) simula$on
• An agent capable of – Reproducing realis$c mobility – Learning of preferences, behaviors (peers modeling) – Elicita$on – Op$miza$on – etc.
The “agent” paradigm
• An agent can operate in two different ways: – Autonomous Interface Agent (AIA) assis$ng & interac$ng with humans
– Autonomous Agent (AA) to simulate the human
The “agent” paradigm
Learning • Behavior • Preferences • Ac$ons Assis$ng
Reported ac$ons, choices, type, preferences
Behaving with a par$cular type, profile, preferences
AIA Real behavior
Simulated behavior
AA
MAS on StarBED stack
プロバイダ網
Wireless mesh . . . . . Pluggable
Net Emula+on
PC Sensor
Nodes Emula+on
Web Twider
. . . . . Pluggable Wired (Ether)
Wireless (LTE)
Physical Layers
Satellite com
Car
Wireless (Wi-‐Fi)
IP Phone Embedded Navi app
StarBED
Android
Mobile app
MAS on StarBED stack
Disaster scenario generator
People behavior
Physical state
Failures, etc.
プロバイダ網
Wireless mesh . . . . . Pluggable
Net Emula+on
PC Sensor
Nodes Emula+on
Web Twider
. . . . . Pluggable Wired (Ether)
Wireless (LTE)
Physical Layers
Physical Simula$on (Terrain data)
Determina+on of the affected areas
Satellite com
Car
Wireless (Wi-‐Fi)
IP Phone Embedded Navi app
StarBED
Android
Mobile app
scenario
MAS on StarBED stack
Disaster scenario generator
People behavior
Physical state
Failures, etc.
プロバイダ網
Wireless mesh . . . . . Pluggable
Net Emula+on
PC Sensor
Nodes Emula+on
Web Twider
. . . . . Pluggable Wired (Ether)
Wireless (LTE)
Physical Layers
Physical Simula$on (Terrain data)
Determina+on of the affected areas
Agent θ3 . . . . . Pluggable
Agent
Rou$ng . . . . . Pluggable
MAS
Mechanism Design Conges$on
Satellite com
Car
Wireless (Wi-‐Fi)
Agent θ2
Coordina$on
IP Phone Embedded Navi app
StarBED
Behavioral & Economic type
Macroscopic & Organiza+onal type
§ Strategic types? § Risk aversion? § Uncertainty aversion? § Preferences/U$lity? § Coopera$ve behavior?
Encounter Protocols
Agent θ1
Android
Mobile app
scenario
MAS on StarBED stack
Disaster scenario generator
People behavior
Physical state
Failures, etc.
プロバイダ網
Wireless mesh . . . . . Pluggable
Net Emula+on
PC Sensor
Nodes Emula+on
Web Twider
. . . . . Pluggable Wired (Ether)
Wireless (LTE)
Physical Layers
Physical Simula$on (Terrain data)
Determina+on of the affected areas
Agent θ3 . . . . . Pluggable
Agent
Rou$ng . . . . . Pluggable
MAS
Mechanism Design Conges$on
Satellite com
Car
Wireless (Wi-‐Fi)
Agent θ2
Coordina$on
IP Phone Embedded Navi app
StarBED
Behavioral & Economic type
Macroscopic & Organiza+onal type
§ Strategic types? § Risk aversion? § Uncertainty aversion? § Preferences/U$lity? § Coopera$ve behavior?
Encounter Protocols
Agent θ1
Android
Mobile app
scenario
MAS on StarBED stack
Behavioral models β1 ...βn emula$ng the mobility of the drivers, or cars, which is different from an AIA assis$ng the drivers
Example: Emula$ng a traffic coordina$on task in a disaster scenario (congested) involving car-‐embedded & mobile app agents represen$ng risk seeking, uncertainty averse drivers (only α% are coopera$ve). The coordina$on mechanism uses a sensor network
Experiment Network
Boo$ng instruc$ons Configura$on data Agent stuff, media$on
Agent Agent
(x, y) at t
Visualizer
Qomet Qomet
Experiment Node Experiment Node
(x, y) at t Agent stuff
Simula$on Manager or
MAS on StarBED stack
Experiment Network
Boo$ng instruc$ons Configura$on data Agent stuff, media$on
Agent Agent
(x, y) at t
Visualizer
Qomet Qomet
Experiment Node Experiment Node
(x, y) at t Agent stuff
Simula$on Manager or
MAS on StarBED stack
Agent
OSM DM
MAP
BM Visualizer
Server
OSM DM
MAP
Server
Traffic simula$on
• Agents will have to emulate drivers behavioral models
• Dynamic and interac$ve mobility genera$on based on 3 forces – Speed v – Accelera$on a – Breaking b
• Interac$on with external forces – Driver ac$ons, through the steering wheel – Environment: traffic lights, routes – Interac$on with other vehicles
Traffic simula$on
Speed difference: Example of rules:
δv!< −3
a!= 0
b!= δv!*urgency
δv!> 3
a!= δv!*urgency
b!= 0
objv!= 0
a!= 0
b!= −v!
δv!= objv
!− v!
Speed modula$on
Default speed, or “desired” speed
v!
a!
b!
objv!
Current speed, resul$ng from the interac$on with the environment
Traffic simula$on
Speed difference: Example of rules:
δv!< −3
a!= 0
b!= δv!*urgency
δv!> 3
a!= δv!*urgency
b!= 0
objv!= 0
a!= 0
b!= −v!
δv!= objv
!− v!
Speed modula$on
Default speed, or “desired” speed
v!
a!
b!
objv!
is based on the steering wheel posi$on and how it changes given any change in the overall direc$on of the vehicle and its velocity angle (θ) Update rule
Angle modula$on
Current speed, resul$ng from the interac$on with the environment
δθ = objd!"−θ
Vsw =Vswt −Vswt−1
Vsw =Vswt−1 +δθ
Default direc$on
Steering wheel value at t-‐1
-‐1 1
Vsw
• Rule-‐based direc$ves or programmable goal-‐oriented agents
Traffic simula$on
Traffic simula$on Agent-‐based, microscopic prototype (before distribu$on)
Traffic simula$on Agent-‐based, distributed, macroscopic prototype
Thank you