Transcript sol05-10

Agents for collaboration in coalition environments
The Intelligent Software Agents Lab
Katia Sycara, Principal Investigator
Ready-to-use software-integration-web technologies
Agent Architecture
Four parallel threads:
• Communicator
• for conversing with
other agents
• Planner
• matches “sensory”
input and “beliefs” to
possible plan actions
• Scheduler
• schedules “enabled”
plans for execution
• Execution Monitor
• executes scheduled plan
• swaps-out plans for
those with higher
priorities
Information Fusion for Command and Control: from Data to Actionable Knowledge and
Decision
CoABS:
Effective Coordination of Multiple Intelligent Agents for Command and
Control
AFOSR PRET F49640-01-1-0542
MAS Infrastructure
MAS Interoperation
Interoperation
Translation Services Interoperator Services
Interoperation Modules
Capability to Agent Mapping
Capability to Agent Mapping
Middle Agents
Middle Agent Components
Name to Location Mapping
Name to Location Mapping
Agent Name Service
ANS Component
Security
Security
Certificate Authority Cryptographic Service
Security Module
Performance Services
MAS Monitoring Reputation Services
Performance Service Modules
Multi-Agent Management Services
Management Services
Logging Activity Visualization Launching
Logging and Visualization Components
ACL Infrastructure
MAS
Infrastructure
Public
Ontology
Protocol Servers
Communications Infrastructure
ACL Infrastructure
Individual
Infrastructure
Parser, PrivateAgent
Ontology,
Protocol Engine
Communication Modules
Discovery
Technical Approach:
Develop new methods for high level information fusion
Level 2: force recognition (recognizing groups)
Level 3: inferring intent and threat
Level 4: identifying & acquiring needed information
•Developing simulation, display, and infrastructure for human-system interaction research
•Conducting verification and validation studies with human users
Automatically generated by CMU’s terrain analysis
software
Subject Matter Expert’s MCOO (Modified
Combined Obstacle Overlay)
Adjustable Autonomy for the Battlefield
Cooperative Attack Realtime Assessment (CAMRA)
Operational Capability:
We are developing and testing
search munition control
strategies using both a high
fidelity 6-dof simulation of the
LOCAAS and medium fidelity
6-dof simulation of an
unspecified search munition.
We are adapting team oriented
programming approaches to
provide sophisticated
planning and cooperation
capabilities to teams of
munitions.
CaveUT
Convergent Tools
Private/Public Keys
Performance Services
Discovery Message Transfer
Operational Capability:
We have developed a suite of interacting tools using the
OTB military simulation and the Unreal engine that
allow us to simulate the warfighter’s environment
anywhere on the battlefield. By combining ISR data,
human communications, and realistic tasks we can test
and evaluate conops and technologies for network centric
warfare. Without the complexity allowed by these
networked tools it would be impossible to test our
research hypotheses involving active annunciation and
information filtering and distribution.
Simulation tools developed in this project have already
been transitioned to AFRL and ARL laboratories and are
in use at universities here and in Europe.
Munition
simulation
OTB
simulation
Message Transfer Modules
Robotic control
interface
Operating Environment
Machines, OS, Network, Multicast Transport Layer, TCP/IP, Wireless, Infrared, SSL
Functional Architecture
Terrain
analysis
Unreal
engine
Technical Approach:
Agents construct and evaluate plans based on multi-dimensional effects and interactions among effects.
We are developing techniques to allow wide area search munitions to cooperate in order to locate and
attack targets, perform battle damage assessment, etc.
Control concepts and prototype interfaces to allow humans to control and monitor cooperating search
munitions
USAR
Urban Search and Rescue
1. Hardwired Agent Communications
2. ANS Location
Registry
“I know you, the service you provide, and where you are located.”
The RETSINA MAS Provides:
• Dynamic team coordination, supporting teamwork between entities of varying
capabilities;
• Adjustable Autonomy for adaptively sharing control, responsibilities, and commitments
at all task abstraction levels and by all types of team members (agents, robots, people);
• Abstraction-based tiered robot architecture that consists of incremental functional
abstractions with real-time behavior based controllers at the lowest level, executive nearterm explicit reasoning and scheduling at the middle level, and declarative planning and
communication at the top level;
• Scalability to larger or smaller numbers of robotic and software agents without affecting
the team goal through loss of coordination, etc.
Search and Rescue Results:
“I know who I want to speak with, I just need to find them. The agent I am looking for
is in my local domain.”
In RETSINA, agents known to each other do not need centralized intermediaries
to communicate.
Agent Name Server ANS)
3. ANS Hierarchy Partners
“The agent I am looking for may not be in my local domain, so I will check with
the ANS hierarchy partners with whom I am familiar. My partners will forward my
The ANS is a server that acts as a registry or “white pages” of agents, storing agent
request to their known partners, who will search their directories for
names, host machines, and port numbers in its cache. The ANS helps to manage
me.” Dynamic
Software
Robotics Inst.
Engineering
4. Multicast
Discovery--works
besta mechanism
in Local Area
Networks
(LANs)
inter-agent
communication
by providing
for locating
agents.
Discovery of
infrastructure
and agent
CMU
services Agents
Computer Services
Group
Engli sh Dept.
New Middle-A gents
are discovered
across WAN, and the
query propagates
within the LANs.
“Hello, my name is Agent B and my location is Y.”
“Hello, My name is Mike and
my location is
orion.andrew.cmu.edu and I
can perform X task.”
Multicast
Discovery
Interface
PER
ANS
Image-Rec
not found.
Matchmaker
ANS
NAS A
Project Goal:
Govt. Agencie s
Lookup
Image-Rec
Agent
agents
Image-Rec
found!
5. Agent-to-Agent (A2A)
“I want to find agents, services and infrastructure beyond my LAN. I don’t
know who or where these entities are, but I need to look for them across a
Wide Area Network (WAN).”
RETSINA A2A technology uses the existing Gnutella P2P network to
gather information about agents, services and infrastructure components
so that agents may connect across WANs to access each other’s
services.
ANS
Matchmaker
“Hello, My name is Joe and
my location is
areolis.cimds.ri.cmu.edu and I
can perform Y task.”
With Multicast Discovery, agent registrations, locations and capabilities are “pushed”
to other agents and infrastructure components, which discover each other and avail
themselves to each other’s services.
To develop hybrid teams of autonomous
heterogeneous agents—including cyber
agents, robots, and humans—that
intelligently coordinate and plan to
accomplish urban search and rescue in
disaster situations. We envision a MultiAgent System (MAS) in which humans,
agents, and robots work together
seamlessly to provide aid as quickly and
safely as possible in the event of an
urban disaster.
Human
Environment: disaster area
Corky
http://www-2.cs.cmu.edu/~softagents/project_grants_NSF.html