Validation of Autonomous Concepts using the ATHENA Environment

1
Validation of Autonomous Concepts using the
ATHENA Environment

• ESA On-Board Autonomy workshop
• ESTEC, Noordwijk, The Netherlands, October 2001

guettier_at_parc.xerox.com bruno.patin_at_dassault-aviat
ion.fr jean-francois.tilman_at_axlog.fr
2
Agenda

• Presentation
• Models used for realistic representations
• Architecture of the simulator
• Description of a scenario
• Two autonomy experiments in aeronautic and space
  domains
• Conclusion

3
Specifications

• Dassault Aviation studies autonomy techniques for
  Unmanned Aerial Vehicles (UAVs)
• Need for a tool for rapid prototyping and
  validation of autonomous behaviours of
  distributed systems in complex environments
• Existent code has to be reused in the simulation
  without any reengineering
• The tool has to be generic enough to allow easy
  enrichments.
• Athena

4
Overview of Athena

• Athena a distributed simulation framework for
  autonomous behaviours
• Rapid design and prototyping of heterogeneous
  distributed architectures in complex environments
• Simulation of their behaviour close to the final
  system
• Because of the genericity, the user can choose
  the abstraction level of his simulation

5
Validation of autonomous behaviours
Problem specification
Environment description in Athena
ConceptionPrototyping
Simulation
Integration of developed code at any granularity
level
Development of autonomy functions
Assessment,Comparisons with the foreseen
behaviour
6
Domain modelling

• Needed models
• realistic environment representation
• on-board processing power
• agent models
• three layers
• physical layer
• system layer
• agent layer

agents
on-board systems
physical world
7
Physical layer

• Representation of complex environments in the
  physical world
• Continuous domain
• parameter data container(position, signal
  amplitude, etc.)
• interaction computation of new parameter
  values.(computation of an estimated position )
• Discrete domain
• state state of a complex object
• transition changing of states
• event activation of a transition

x1
D
diff
x2
condition
automaton
on
stop
ok
init
off
start
8
System layer

• When simulating a complex system, we must
  introduce on-board calculators.
• task
• represents a real-time task
• periodically or sporadically activated
• performs user calculations on parameters
• take into account an execution time
• process
• represents a set of scheduled tasks

Process
task
task
task
9
Agent layer

• A multi-agent system contains agentswhich can
• perceive new facts to update their knowledge
  base
• interact with each other
• reason and take decisions.
• The agent layer is based on the lower layers to
  define
• specific data types
• perception and reasoning functions

search for a path
10
Simulator architecture

• Distribution over anheterogeneous network
• one or several servers
• manage data and calculations
• clients
• read and interact with data
• connected through an Object Request Broker (ORB)
• Synchronisation
• distributed clock

Server 2
Server 1
automata parameters interactions processes
automata parameters interactions processes
sequencer
sequencer
synchroniser
synchroniser
ORB
corba naming service
client
11
Processing tool

• Objectives
• introduction of processingon large amount of
  data
• easy composition ofthis processing
• computation on remote stations withspecific
  features
• connection to other computation softwareprograms
• Useful to supply code to interactions and tasks

ORB
Processingserver
Simulation server
composite
interaction
processing
task
composite
processing
processing
SciLab
Prolog
12
Simulation design
PROTOTYPE Aircraft PARAMETER Long x 5
PARAMETER Long y 5 END

• An architecture description language (ADL) has
  been designed for simulation needs.
• Enables object-oriented descrip-tions of the
  simulated architectures
• composition
• inheritance
• allows use of user-specific datatypes and
  functions
• reusability of prototypes

PROTOTYPE Plane IS Aircraft PARAMETER Long
pilot 1 END
1
PROTOTYPE Ucav IS Aircraft END
2
PROTOTYPE Formation INSTANCE Plane leader
INSTANCE Ucav wingman1 INSTANCE Ucav
wingman2 END
Extracts from the example description
13
Aeronautic example (1)

• Based on a real study by Dassault,to evaluate
  the UCAVs autonomy, especially the planning
  techniques
• Description
• An autonomous formation must fly through the
  enemy territory and avoid threats.
• Enemy radars detect planes, track them, and
  launch missiles
• When the formation detects a radar
• the leader replans its flightplan and
  communicates it to the wingmen
• the wingmen compute their own flightplan

The Dassault Aviation Petit Duc UAV
14
Aeronautic example (2)

• Introduction of specific types
• coordinates, attitudes
• graph
• flightplan
• Introduction of user functions
• flight laws
• electro-magnetic propagation laws
• planning functions which are tested

15
Aeronautic example (3)
- Enemy radars - tracking radars - acquisition
radars - on - off - Graph determining
the possible flightplans - Autonomous formation -
1 leader - 2 wingmen
16
Aeronautic example (3)

• The detection radars switch on.
• They are detected by the planes.

17
Aeronautic example (3)

• The detection radars switch on.
• They are detected by the planes.
• The leader centralises detection information from
  the formation
• It evaluates radar positions.
• It replans a new flightplan.

18
Aeronautic example (3)

• The detection radars switch on.
• They are detected by the planes.
• The leader centralises detection information from
  the formation
• It evaluates radar positions.
• It replans a new flightplan.
• The wingmen receive the leader flightplan.
• They compute their own new one.

19
Aeronautic example (4)

• After simulation
• comparison of the results with foreseen results
• explanation of the autonomous behaviour
• some traces are available to help in this work
• Beginning of a new cycle with
• refinement of the scenario
• correction of the autonomy functions
• new simulation

20
Space example (1)

• A formation of spacecraft must realize
  observations
• A leader is elected
• It coordinates the formation, computes mission
  plans, schedules ground observation requests
• Spacecraft extract and execute their sub-plans,
  and realizing observations, communications and
  data upload.
• Drifts and errors in plan realisation are
  communicated through agent system.
• Master triggers replanning when necessary

21
Space example (2)

• During the simulation
• visualisation of the values of the parameters,
  the state of the satellites, their paths
• traces for futureexploitations
• After simulation
• analysis of results

22
Related works

• CADCOM specialised for underwater simulations
  with real or virtual AUVs.
• Real Time Mission Lab (Georgia Tech, Aviation and
  Missiles Command, Honeywell)Reco UAV and Ground
  Platoons
• MUSE (JTC/SIL) UAV Assessment
• MISURE European project to study autonomous
  formation

23
Conclusion

• Athena enables prototyping and validation without
  specific simulation developments
• Validated onto a realistic autonomy scenario by
  Dassault Aviation
• Usable for other purpose than autonomy
• Further works
• HLA interfaces
• integration of Esterel
• intuitive configuration and control man-machine
  interface

24
Future of Athena

• A club for Athena partners and users
• (simulation kernel, ADL)
• (management, aeronautic domain)
• (3D visualisation, configuration)
• More information
• http//www.axlog.fr/R_d/athena/athena_en.html
• http//www.prolexia.fr/francais/ingenierie/referen
  ces/Athena_doc.htm