2008 01 research overview.ppt · · 2009-09-29gracefully handles GPS outages ... R. Beard...

Vehicle TestbedAi fAirframes

3-Axis Angular Rate & Acceleration Measurement20 Point Sensor Temperature CompensationKalman Filter Attitude EstimationOptional “piggy-back” Modem

Autopilot

Fixed Wing

Configurable Failsafes2-Axis Magnetometer2-Axis Gimbal SupportDead reckoning filter gracefully handles GPS outagesMultiple-UAVs

Gimbaled Camera SoftwareHovercraft Tailsitter

Smart LoitersAuto-Trim

16.7 grams

2” x 1.37”x .47”

Gimbaled Camera SoftwareBYU DesignDriven by Hitec Servos

Azimuth: HS-77BB Low ProfileTorque: 61.1oz-in

Elevation: HS-50 FeatherKestrel AutopilotKestrel Autopilot Image Tracking

SoftwareImage Tracking

SoftwareAviones

MAVCamera

Torque: 8.33 oz-inPWM Input Signal (900-2100 μs)

Camera: Panasonic KX-141Resolution: 480 lines Sensitivity: 2 lux City Maker

DEM Map

City Maker

DEM Map

Research Overview Guidance, Navigation, Vehicle ControlR. Beard

FOV lenses available: 19°, 30°, 64°, 120°Format: NTSC (PAL available)

Total Weight: 7.4 ounces Virtual CockpitVirtual Cockpit Path PlannerPath Planner

TailsitterObjective: Design guidance, navigation, and control strategies for tailsitter UAVs. Challenges include attitude estimation, and transitions between hover and level flight.

P i i l I ti t Ti M L i R d B d

Problem Summary1: VTOL MAV Aerodynamic Modeling: Develop a preliminary design tool for low Reynolds number aircraft via a modified numerical lifting line algorithm.

2 Ad ti A t il t f VTOL MAV D lPrinciple Investigators: Tim McLain, Randy Beard

Sample Publication: Nathan B. Knoebel, Stephen R. Osborne, Deryl O. Snyder, Timothy W. McLain, Randal W. Beard, Andrew M. Eldredge “Preliminary Modeling, Control, and Trajectory Design for Miniature Autonomous Tailsitters,” AIAA Conference on G id N i ti d C t l K t CO 2006

2: Adaptive Autopilot for VTOL MAVs Develop an adaptive autopilot capable of control in both hover and cruise flight modes.

3: Enhanced Attitude Estimation Develop and implement attitude estimation algorithms capable of largeGuidance,Navigation, and Control, Keystone CO, 2006, paper no.

AIAA-2006-6713.

Funding Source: AFRL/MN.

Approach Results

implement attitude estimation algorithms capable of large attitude angles and high-performance maneuvers that are suitable for the sensors and processors used on MAVs.

Approach Results1. Represent attitude with unit

quaternion

2. Derived a novel extended Kalman

Successful flight test in 2005-2007.

Hover to level transition.

filter for attitude estimation.

3. Developed inner feedback loops based on adaptive control around unit quaternion.

Level to hover transition.

Long term hover.

Autonomous take-off and


4. Developed adaptive speed controller for power control.

Autonomous take off and landing. (Landing is still rough.)

Path PlanningObjective: Develop path planning techniques for UAVs that can be implemented in real-time on simple computational hardware.

Principle Investigators: Randy Beard, Tim McLain

Problem SummaryGiven the current position of the UAV, a desired goal configuration, and a terrain map, plan a feasible path through the terrain.

Sample Publication: Stephen Griffiths, Jeff Saunders, Andrew Curtis, Blake Barber, Tim McLain, Randy Beard, “Obstacle and Terrain Avoidance for Miniature Aerial Vehicles,” Chapter 7 in Advances in Unmanned Aerial Vehicles: State of the art and the road to autonomy

g

Waypoint path planner

Path manager

Vector field path following

List of waypoints

Tracking error

Status

Destination,obstacles

Hierarchical decomposition of the problem into waypoint path Vehicles: State of the art and the road to autonomy,

edited by Kimon P. Valavanis, Springer Verlag, 2007.

Funding Source: AFOSR, AFRL/MN.

Approach Results

Autopilot

Micro air vehicleOn-board sensors

Position errorfrom camera

Servo commandsState estimatorState estimator

Wind

yp pplanning and dynamic path following.

Approach Results

Developed path planning algorithms that use b th th V i

First successful flight test in 2005.

Successful flight through rural canyon in central Utah.

Successful flight through simulated urban terrainboth the Voronoi algorithm and the Rapidly Exploring Random Tree (RRT) algorithm.

Successful flight through simulated urban terrain


Trajectory Following

Objective: Given a desired orbit or waypoint path, precisely follow the path even in heavy winds.

Principle Investigators: Tim McLain, Randy Beard

Vector Field MethodTwo key characteristics:

Path following instead of trajectory trackingp g , y

Sample Publication: Derek R. Nelson, Blake Barber, Timothy W. McLain, Randal W. Beard, “Vector Field Path Following for Miniature Air Vehicles,” IEEE Transactions on Robotics, vol. 23, no. 3, June, 2007, p 519-529.

Be “on the path” vs. “follow the rabbit”

Controlling course instead of heading

Funding Source: AFOSR, AFRL/MN.

Key AccomplishmentsFlight Test Results Key AccomplishmentsFlight Test ResultsFirst successful flight test in 2003.

Robustness to wind disturbances

Avg tracking error less than 3 wingspans

Hitting the waypoint Equalized path length Circular orbit

Avg tracking error less than 3 wingspans when wind is 30 to 50% of commanded air speed

Licensed to Procerus Technologies in 2004 as

Equalized path length Simulated urban terrain


gpart of the Kestrel autopilot system.

Hundreds of hours of successful flight tests.

Laser Range Finder

Objective: Use a forward pointing laser range finder to detect and avoid obstacles.


Laser Range FinderOpti-Logic RS400 Laser Rangefinder

400 tSample Publication: Stephen Griffths, Jeff Saunders, Andrew Curtis, D. Blake Barber, Timothy W. McLain, Randal W. Beard, “Maximizing miniature aerial vehicles,” IEEE Robotics and Automation Magazine, vol. 13, no. 3, 2006 p 34-43

400 meter range

10 Hertz update rate

6 oz2006 p. 34 43.

Funding Source: AFOSR.

Approach Results

6 oz

1.8 Watts

Approach Results1. While following a nominal waypoint path, laser

detects object.2. Upon detection, insert cylindrical object into world

map and plan path around the obsticle.

First successful flight test in 2005.

Detected and avoided a 50 meter hight

3. Continue until the obstacle has been avoided. g

building on the BYU Campus.

Original Waypoint Path

UnexpectedObstacle


Optic Flow Sensor

Objective: Using optic flow sensors, detect and track the walls of a rural or urban terrain.


Optic Flow Sensor• Computationally cheap • Light weight – 0.8 oz (22 grams)

Small form factor 1” 1”Sample Publication: Stephen Griffths, Jeff Saunders, Andrew Curtis, D. Blake Barber, Timothy W. McLain, Randal W. Beard, “Maximizing miniature aerial vehicles,” IEEE Robotics and Automation Magazine, vol. 13, no. 3, 2006 p 34-43

• Small form factor – 1” x 1” x 1.5” • Inexpensive – $80 including optics• Fast – computes optic flow at 2300 fps2006 p. 34 43.


Approach Results

at 2300 fps• Range: 80 m using 1.2 deg FOV lens. • Custom design at BYU

Approach ResultsFirst successful flight test

in 2005.

Successful terrain following (height above ground) in 2005.

Flew through a rural canyon in 2005


canyon in 2005.

Kestrel AutopilotThe Kestrel autopilot began in 2001 as a mobile

robot control board for a robot soccer senior design project. The board was adapted for flight control in 2002 and successfully flew its first small UAV in 2003 In 2004 the technology was licensed to

3-Axis Angular Rate & Acceleration Measurement20 Point Sensor Temperature CompensationKalman Filter Attitude EstimationOptional “piggy-back” ModemConfigurable Failsafes2003. In 2004 the technology was licensed to

Procerus Technologies, a new start-up, which has been successfully commercialized the product and now employs ten engineers.

T k l d h K l il i h

2-Axis Magnetometer2-Axis Gimbal SupportDead reckoning filter gracefully handles GPS outagesMultiple-UAVsSmart LoitersTo our knowledge, the Kestrel autopilot is the

smallest available fully functional autopilot in the world.

Flowchart Technology Transition

Smart LoitersAuto-Trim

16.7 grams

2” x 1.37”x .47”

Flowchart Technology TransitionUniversity AFRL

Autopilot for BATCAM Program (2003)

FundingUAV

On-boardcamera

Sensors:GPS, gyros, etc.

Actuators

Autopilot

• Servo loops• Attitude est.• Waypoint nav

Autopilot

• Servo loops• Attitude est.• Waypoint nav Gimbal

Helios FPGA Board:Color / Feature tracking

Avionics for BATCAM (2004)

Army TACMAV (2005)AF Night Hawk (2006)

Kestrel Autopilot (2004)Flight Control Software (2004)

Video stabilization (2006)Target tracking (2006)Target localization (2006)

Sensor platform (2006)Unique gimbal design (2006)

Funding

By-pass switch

900 MHztransceiver900 MHz

transceiver2.4 GHz

transmitter2.4 GHz

transmitter

RC Controls

Laptop• ground station

- telemetryi t l

Laptop• ground station

- telemetryi t l

900 MHztransceiver900 MHz

transceiver

2.4 GHztransmitter2.4 GHz

transmitterserial


Industry 2004 Start-up2004 Start-up2006 Start-up2006 Start-up

Ground

Interface DevicesPDA & Voice

- waypoint planner- human interface

• Path planning• Video display.• Video processing

- waypoint planner- human interface

• Path planning• Video display.• Video processing

frame grabber

802.11b

Target GeolocationObjective: Allow a user watching a video stream to click on a target of interest, and to enable the system to continuously track the target and accurately estimate its GPS coordinates.

Principle Investigators: Tim McLain Randy Beard

Problem SummaryStep 1. User identifies target of interest in the video stream. Step 2. Computer vision algorithm tracks the target in the video stream.Principle Investigators: Tim McLain, Randy Beard,

Clark Taylor

Sample Publication: D. Blake Barber, Joshua D. Redding, Timothy W. McLain, Randal W. Beard, Clark N. Taylor, “Vision-based Target Geo-location using a Fixed-wing Miniature Air V hi l ” J l f I t lli t d R b ti S t l 47 4

the video stream. Step 3. UAV sensors are used to estimate the relative position of the target, and its GPS coordinate is estimated from the UAVs GPS sensor.Step 4. Servo pan-tilt gimbal so that

Vehicle,” Journal of Intelligent and Robotic Systems, vol. 47, no. 4, December, 2006, p. 361-382.


Approach Results

object is in center of imageStep 5. Modify flight path of UAV to orbit object

Approach Results1. Kalman filter raw data

2. On-board gimbal calibration

3 Optimal flight path selection 20

40

60

80MAV loctarget esttarget est avg

Relative target estimation error less than 2 meters after one orbit

First successful flight test3. Optimal flight path selection

4. Wind estimation andcompensation

5. Telemetry Synchronization -80

-60

-40

-20

0

20 First successful flight test in 2005.

Licensed to Procerus Technologies in 2006.


5. Telemetry Synchronization

6. Range Estimation Using Digital Elevation Models

-80 -60 -40 -20 0 20 40 60 80-80

target estimate error = 1.9611m

Commercial product: OnPoint

Target ProsecutionObjective: Use computer vision to autonomously land a small fixed wing UAV on a visually distinct target.

Principle Investigators: Randy Beard, Tim McLain, Jim Archibald

Problem SummaryThere are many applications that require precision landing for micro air vehicles. These include:• Convoy support – landing in the back of a vehicle.

Sample Publication: D. Blake Barber, Stephen Griffiths, Timothy W. McLain, Randal W. Beard, “Autonomous Landing of Miniature Aerial Vehicles,” AIAA Journal of Aerospace Computing, Information, and C i ti l 4 5 M 2007 770 784

y pp g• Micro munitions.

A user is often tasked to assist the vehicle in landing.

Computer vision and missile guidance algorithms can be d t i l diCommunication, vol. 4, no. 5 May, 2007, p. 770-784.


Approach Results

used to ensure very precise landings.

Approach ResultsFirst successful flight test in 2006.

Landed in the back of a moving truck in 2006

Adapted Proportional Navigation algorithm to bank-to-turn UAVs:

Step 1. User identifies location to land in video stream.Step 2 Computer vision tracks the target in the video truck in 2006.

Licensed to Procerus Technologies in 2007.

Currently a commercial product.

Step 2. Computer vision tracks the target in the video.Step 3. Ego motion is estimated using IMU and removed from relative position vector.Step 4. Desired acceleration is computed in the body frame.Step 5. Acceleration is converted to commanded roll and


y pStep 5. Acceleration is converted to commanded roll and pitch angles.

Road Following

Objective: Use computer vision to autonomously track a perimeter on the ground.

Principle Investigators: Randy Beard, Clark Taylor

Computer Vision

• Threshold in Hue-Saturation-Value (HSV) color spacep g y y

Sample Publication: Joseph Egbert, Randal W. Beard, “Road Following Control Constraints for Low Altitude Miniature Air Vehicles,” American Control Conference, New York, New York, July 2007, p. 353-358.

Value (HSV) color space.

• Connect the classified pixels into components

• Find the largest components

Funding Source: NASA.

Approach Results

Find the largest components and return the top center pixel.

Approach ResultsControl a desired heading rate for a skid-to-turn platform equipped with a strapped-down camera: Image Processing

Camera

Captured Image

rs)

100

150

200

250• Successful flight test in 2007.

• Used in conjunction with

Control a desired roll angle and a gimbal elevation angle for a bank-to-turn platform equipped with a gimbaled

Guidance and Control

Road Pixel Location (εx,εy)

PN Control Law

Relative East (meters)

Rel

ativ

e N

orth

(met

er

-300 -200 -100 0 100 200 300-250

-200

-150

-100

-50

0

50cooperative perimeter tracking and image mosiacing.

• Rural road successfully tracked for over two miles.


q pp gcamera:

UAS and Autopilot Telemetry

φ, αel

Cooperative Rendezvous

Objective: To maximize the element of surprise, a team of UAVs is autonomously maneuvered to simultaneously arrive at a specific location. no-flydetection

regionno-flySAMsite

Problem Summary


Sample Publication: Timothy W. McLain, Randal W. Beard, “Coordination Variables, Coordination Functions, and Cooperative Timing Missions,” AIAA Journal of Guidance, Control, & Dynamics, vol. 28, no. 1., January, 2005, p. 150-161. boundary

boundary

boundaryloiterpenalty

Wind25knots

X

threatthreat

Funding Source: AFOSR. 25 miles no-fly

Approach Results

J(x,u)

Cost vs timingJ(x,u)

Cost vs timing

#3

[ ] ] ][[

Path #1 Path #2

Path #3Θ(x) Set of feasible CVs

variable velocity yields range

[ ] ] ][[

Path #1 Path #2

Path #3Θ(x) Set of feasible CVs

variable velocity yields range

Approach Results

• Successful flight test with three UAVs in summer 2005.

[ ] ] ][[

Path #1 Path #2

Path #3

θ

Cost vs. timing

[ ] ] ][[

Path #1 Path #2

Path #3

θ

Cost vs. timing

J(x,u) φ (x, θ )i Pseudo-inverse of fselects lowest cost

J(x,u) φ (x, θ )i Pseudo-inverse of fselects lowest cost#1

#2 • Simultaneous rendezvous in moderate winds.

• Timing error less than 0.1 seconds.


[[ ]] ]] ]][[[[

Path #1 Path #2

Path #3

θ

[[[[ ]] ]][[[ ]][[ ]] ] ]][[[ [ ]

selects lowest cost path for a given θ

[[ ]] ]] ]][[[[

Path #1 Path #2

Path #3

θ

[[[[ ]] ]][[[ ]][[ ]] ] ]][[[ [ ]

selects lowest cost path for a given θ

#1 seconds.

Cooperative Perimeter MonitoringObjective: Use a team of cooperatively monitor a fixed or dynamically changing border.


Problem SummaryA team of MAVs with flight durations of 1-2 hours, are tasked to cooperative monitor a perimeter. The communications range is limited and precludes constant communication. The perimeter may be changing as

Sample Publication: David W. Casbeer, Derek B. Kingston, Randal W. Beard, Timothy W. McLain, Sai-Ming Li, Raman Mehra, “Cooperative Forest Fire Surveillance Using a Team of Small Unmanned Air Vehicles,” International Journal of Systems Science, vol.

communication. The perimeter may be changing as would be the case for a forest fire perimeter.

37, no. 6, May, 2006, p. 351-360.

Funding Source: NASA.

Approach Results

Possible applications include border patrol, fire surveillance, hazardous material monitoring

Approach Results

if rendezvous with neighbor then Theorem

Simple decentralized algorithm. Guaranteed optimal performance.

• Successfully flight tested in 2006-2007 using three MAVs.• Robust with respect to wind.• Seamlessly handles vehicle insertions and deletions.• Works for arbitrarily small communication radius.

if rendezvous with neighbor then- calculate shared border position- travel with neighbor to shared border- set direction to monitor own segment

else if reached perimeter endpoint- reverse direction

else

If the perimeter and number of agents are fixed, then consensus is achieved in 3T, where T is the time required for one MAV to traverse the perimeter.


else- continue in current direction

perimeter.Proof:

Based on consensus ideas.Step Change Sinusoidal Change

Cooperative Consensus SeekingObjective: Facilitate the development of distributedObjective: Facilitate the development of distributed cooperative control algorithms for teams of UAVs that must operate in environments where the communication topology is sparse, unreliable, and constantly changing, and where communication packets are frequently lost.

Problem Summary

Input Data Output DataCoordinationAlgorithm

Vehicle i

Input Data Output DataCoordinationAlgorithmInput DataInput Data Output DataOutput DataCoordinationAlgorithm

CoordinationAlgorithm

Vehicle i

Principle Investigators: Randy Beard

Sample Publication: Wei Ren, Randal W. Beard, Distributed Consensus in Multi-Vehicle Cooperative Control, Communication and Control Engineering

Input Data Output DataCoordinationAlgorithm

Vehicle j

Consensus onOutput Data

Consensus onInput Data

Input Data Output DataCoordinationAlgorithmInput DataInput Data Output DataOutput DataCoordinationAlgorithm

CoordinationAlgorithm

Vehicle j

Consensus onOutput Data

Consensus onInput Data

C b f d i h h i f hSeries, Springer Verlag, New York, 2007, ISBN: 978-1-84800-014-8.

Funding Source: AFOSR, NSF.

Approach Results

Consensus can be formed at either the input of thecoordination algorithm, the output, or both.

We will focus on the case where the output data (coordinationvariable) is to be synchronized.

Approach ResultsDeveloped mathematical theory for consensus algorithms over distributed, time-varying, noisy communication channels. Algorithms interface cooperative control algoithms with the communcation network

Successful application in simulation to:

• UAV and spacecraft formation flying

• Mobile ground robotscommuncation network. Mobile ground robots

• Cooperative rendezvous

• Cooperative perimeter surveillance


• Cooperative search

Flight results for cooperative perimeter surveillance demonstrated in 2007.

Cooperative OrbitingObjective:. Keep target in camera view even with possible occlusions while satisfying UAV dynamics

Principle Investigators: Randy Beard

Problem SummaryA team of UAVs, each equipped with a gimbaled camera, are orbiting a moving ground target. The UAV autonomously distributed themselves around an orbit so that sensor occlusions are minimizedPrinciple Investigators: Randy Beard

Sample Publication: Derek B. Kingston, Randal W. Beard, “UAV Splay State Configuration for Moving Targets in Wind,” in Advances in Cooperative Control and Optimization, edited by Michael J. Hirsch, Lecture

that sensor occlusions are minimized.

Notes in Computer Science, Springer Verlag, (to appear).

Funding Source: NSF, AFOSR.

Approach ResultsApproach ResultsHeading command is proportional to (1) distance from orbit, and (2) relative spacing. Essential idea: UAVs increase or decrease radius to achieve spacing.

• Successful flight test with three UAVs in summer 2007.

Salient Features:• Decentralized – Only neighbor information required• Robust to insertion/deletion• Constant UAV velocity

• Can adapt to moving targets.

• Can adapt to heavy winds.


y• Guaranteed to space UAVs equally along the orbit• Minimizes lateral motion of object in image.

2008 01 research overview.ppt · · 2009-09-29gracefully handles GPS outages ... R. Beard...

Documents

Transcript of 2008 01 research overview.ppt · · 2009-09-29gracefully handles GPS outages ... R. Beard...