Control of UAV TeamsPaul Scerri & Katia Sycara Carnegie Mellon University

Michael Lewis University of Pittsburgh

P-LOCAAS Flight Test

AC-130 Flank Support test

Coordinating UAV Teams

1               Acknowledgements

This project is supported by AFRL/MN and involves the contributions of many others including Rob Murphy and Kevin O’Neal from AFRL, Rolan Tapia and Doug Zimmerer of Lockheed-Martin, and Paul Arezina from the University of Pittsburgh

LOCAAS & wide area search munitions

Turbojet

Ladar

30”

40”

AC-130Simulator OTB

WASMSimulator

3D Database

FLIR

/Coord

inate

s

DREN

Entity State, Object State, and

Weapon State PDUs

Fire Control, Redirect

and Superviso

r PDUs– Alert Message– TAI Encountered– ID Target

– Interactive Display– Specify TAI– Fire Authority

OperatorDisplay

3D Database

Video

WASM

dispense

Wide Area Search Munitions (WASMs) are a cross between an unmanned aerial vehicle and a munition. The first of these high concept munitions, the Low Cost Autonomous Attack System (LOCAAS), is envisioned as a miniature, autonomous powered munition capable of broad area search, identification, and destruction of a range of mobile ground targets.

An experimental user interface for controlling WASMs was constructed by adding a toolbar to the FalconView personal flight planning system, a popular flight planning system used by military pilots. The user controls individual or teams of WASMs by sketching ingress paths, search or jettison regions and other spatially meaningful instructions.

FalconView-based Interface

The FalconView interface will be used to launch and direct a live P-LOCAAS prototype that will fly a mission with three simulated teammates. Coordination among munitions could allow WASMs to perform battle damage assessment for one another, stage simultaneous attacks on a target, and perform other coordinated activities that could multiply the effectiveness of such munitions.

An initial evaluation of the FalconView tasking interface was conducted for WASM conops for flank patrol for an AC-130 aircraft supporting special operations forces on the ground. In the test scenarios the WASMs were launched as the AC-130 entered the battlespace. Three scenarios, one training and two with active data collection were flown in an AC-130 simulator by instructors at the Hurlburt Field SOCOM training facility. Controllers were able to effectively direct the munitions and successfully attack a majority of targets.

There are two general classes of robotic coordination, swarms and intentional Swarms have large numbers of homogeneous, low capability individuals who generate intelligentappearing group behavior, but they are incapable of complex coordination involving roles with differentiated behavior. Intentional coordination requires explicit and complex coordination mechanisms, such as reasoning about joint intentions and teamwork. Behaviorally cued swarming is good for formation flying but bad for more variable forms of cooperation such as BDA, joint attacks, flush & hit, etc. that will be needed to make cooperating UAVs a truly effective asset.

Before

AND

Pre-condition

Post-condition

Role WASM Target X destroyed

Terminate Plan

I see a Target at Y

We need someone for a BDA role!

WASM

Designtime

MachinettaMachinetta provides an infrastructure for coordination. Machinetta proxies come with mechanisms for task allocation, commitment and decommitment to plans and other joint activities needed for coordination. Using Machinetta proxies teams of UAVs can behave in an intentional manner to achieve their controller’s objectives

Finding the leversLayers of control

Pre-launch/programming

•Mission planning

•Reactive behaviors agent level

•Team oriented plansIn flight

Parameter tuning: highly nonlinear/unpredictable

–Reactive behaviors (aggressiveness)

–Team plan parameters

Direct command:

•Teleop

•Goals: waypoints, regions, & ROE

Plan-based interaction

–Instantiate team oriented plans

–Fill role in team oriented plans

In order for operators to configure and control teams effectively we are developing methods to create a team performance model to capture the relation between the environment, team configuration parameters and measures of performance. Using the team performance model in reverse allows operators to specify performance tradeoffs and rapidly find a configuration that best meets those constraints. In initial experiments we have demonstrated the ability of an operator to control the global behavior of a large team using a team performance model to guide actions.