mul$agentsimulaon*on* starbed*...wireless* mesh ..... pluggable netemulaon pc sensor nodesemulaon...

Mul$agent  Simula$on  on  StarBED  Deploying  Agent-­‐based  Traffic  Simula$on  on  StarBED  

Rafik  HADFI  <rafik@itolab.nitech.ac.jp>  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