AGENTS for GENERAL AVIATION FREE FLIGHT

1
AGENTS for GENERAL AVIATION FREE FLIGHT

• Primary goal is to facilitate Free
• Flight for General Aviation (GA)
• A new way of managing air traffic
• Departure from the highly structured system to a
  more flexible, distributed system
• Individual aircraft allowed to change route,
  speed and altitude, within limits
• More responsibility for aircraft separation
  safety rested on pilots
• Ultimate decision making authority still lies
  with the air traffic controllers
• Intelligent Agents can help detect and resolve
  conflicts

2
What are agents?

• Essential Characteristics
• Situated (can sense and take actions in dynamic
  environment)
• Goal-oriented
• Autonomous
• Adaptive (learning)
• Social (collaborative)
• Types of Agents (abstract architectures)
• Reactive trigger rules
• Deliberative reasoning, planning
• Cognitive (Mentalistic) beliefs, desires,
  intentions
• Decision theoretic utility functions

3
Agent Architectures

• Production Systems
• Reactive, trigger rules, CLIPS, SOAR
• Search Algorithms A (WX agent)
• Planning Algorithms
• Hierarchical Task Networks (Retsina, TRL)
• Decision Theoretic
• Markov Decision Processes, maximize payoff
• Cognitive (Mentalistic)
• BDI beliefs, desires, intentions
• JACK, PRS, dMARS

4
Roles for Agents in Aviation

• Agents are software processes that can simulate
  decision-making and be adaptive
• In cockpit planning flight path, managing fuel,
  maintaining stability of flight, monitoring
  traffic or weather conflicts
• On ground (TRACON, ARTCC) planning trajectories,
  resolving conflicts, approach metering, handling
  emergencies, coordination with ground ops,
  airlines, etc.

5
Agent Interactions Between Aircraft

• Agents in FMS automatically formulate and
  negotiate En-route and Terminal trajectories
• En route de-crowd routes or traditional jetways
  between fixes (hub-and-spoke), more direct
  flights and custom flight plans, preferences
• Terminal Area efficient sequencing handle speed
  variability better
• Assume communication via ADS-B
• Allow involved aircraft to determine mutually
  acceptable compromise, rather than relying on
  solution by ATC

6

• Conflict detection projected loss-of-separation
  (10-15 minutes out)
• based on current trajectories/flight plans/intent
• range of ADS-B (160nm)
• ATC only has to monitor, or occasionally
  arbitrate
• Decentralized computing reduce bottlenecks and
  decrease sensitivity to failures/attacks
• Off-load approval of minor deviations
• ATC Still maintains ultimate approval authority
• for now, agent solutions will only be advisory,
  minimize disruption of current procedures

7
Collaboration Models

• Consensus Protocols
• Voting/ranking schemes
• Negotiation protocols
• Make incremental concessions until reach
  compromise
• Distributed Constraint Satisfaction
• Global knowledge consistency
• Teamwork
• Hierarchical vs. distributed structures
• Key concepts roles and responsibilities
• Shared plans implicit coordination,
  synchronization
• Theoretical basis Joint Intentions
• (Other methods contract networks, auctions...)

8
Role of Simulated Mental Attitudes

• Intent transmit more than position/vector
• Desire to avoid weather, flight plan, will be
  turning north, descending due to turbulence,
  reason for deviation
• Beliefs
• shared info (weather, congestion, aircraft
  emergencies)
• common picture of situation
• common knowledge STARs, fixes, active runways,
  traffic patterns
• manage uncertainty

9
Concepts for Development of Multi-Agents for Free
Flight

• Strategic (trajectory planning/management) vs.
  Tactical (avoidance maneuvers)
• Actionable decisions
• Alter flight path heading, altitude, speed
• Factors weather, terrain, traffic
• Constraints fuel, speed/alt range
• Preferences time, fuel cost, comfort

10
CDR MODULE
Logic Flow Control for CDR Module
11
Negotiation

• Shared responsibility (with ground too)
• Utility function Flight Plans gt score
• Negotiation by argumentation
• State what is wrong with proposed solution and
  why
• Communicate preferences as well as constraints
• make up when behind schedule
• minimize fuel consumption
• maneuver limitations (safety, comfort)
• Monotonic Concession Protocol (Zlotkin and
  Rosenschein )
• define a finite set of alternative trajectories
• each agent ranks trajectories by utility,
  proposes best
• take turns proposing next best deal till
  utilities match

12
TRL Agents

• Task Representation Language
• Developed at Texas AM Comp. Sci. Dept.
• contact ioerger_at_cs.tamu.edu
• knowledge bases
• declarative rule base (domain knowledge)
• procedural plans/methods for achieving goals
• connection to simulator(s)
• read state information
• trigger actions
• agents can communicate with each other

13
Example Task Description

• (task flight-plan-1 ()
• (method
• (sequence (takeoff KCLL 16)
• (climb-out 3000)
• (turn-heading 350)
• (fly-direct-to KCNW)
• (descend 500)
• (land KCNW 17L))))

command to simulator
invoke sub-task

• Things that can be added
• interaction with ATC (set new way-points,
  altitudes...)
• handling developing weather, emergencies

14
TRL Agent Architecture
15
The SATS Airport Controller

• How to simulate this with agents?
• encode the formal protocol as plans in TRL
• AMM agent - simple task, 1st come-1st serve
• pseudo-ADS-BTCP/IP (must test robustness of
  protocol w.r.t. communications failures)
• test empirically with various scenarios
• arbitrary number of aircraft
• effects of timing, positions, speeds...
• test handoff from ATC, entry to SCA
• arrival/departure frequencies (Poisson distr.)