Distributed Sensing in Horticultural Environments

Distributed Sensing in Horticultural Environments

George KantorCarnegie Mellon University

International Horticultural Congress Lisboa 2010Colloquium 6: Technical Innovation in Horticulture

25 August 2010

Sensor Networks for Agriculture

basestation

node

Sensors(leaf wetness, temperature, humidity, etc.)

field

Internet

• self-contained “nodes” (radio+IO)• ad-hoc network• data collected, relayed back to central

point• can also send control signals

IHC 2010Lisboa25 August 2010

G. KantorCMU Robotics InstituteSensing for Horticulture



Visualizing Time Series(PSU FREC, ZedX Inc.)

FREC Building

North

50m



Visualizing Spatial Variation


G. KantorCMU Robotics Institute

Sensor Net Sensor Requirements• Hands off operation

• No/little calibration required

• Extremely rugged

• Inexpensive

• Generate small amounts of data

• Require low computational power

• Require low electrical power

Examples: today: temperature, RH, PAR, light, rain, soil moisture, soil EC, leaf wetness, wind speed/direction, etc.future: stem water potential, fruit temperature, fruit size, sap flow, others???



Technology Overview: Robot





Laser Scanning



Building Point Clouds





point cloud created by Ben Grocholsky

Technology Overview: Robot

cameras

NDVI



Robot Sensing Requirements• Hands off operation

• Can have non-trivial calibration step

• Moderately rugged

• Can be expensive

• Can generate large amounts of data

• Can require large computing power

• Can require large electrical power

Examples: today: laser scanners, cameras, hyperspectral imagery future: gas exchange, chlorophyll, pheromone, leaf area,…



Robots vs. Sensor Nets

• High spatial resolution

• Low temporal resolution

• Sophisticated sensing

• More Expensive

• Moderate spatial resolution

• High temporal resolution

• Simple sensing

• Less expensive



Robots vs. Sensor Nets




• More Expensive



• Simple sensing

• Less expensive

x



Robots living together in harmony with Sensor Nets




• More Expensive



• Simple sensing

• Less expensive



Information is Worthless…



Information is Worthless…

…unless you use it to do something!



Set Point Irrigation

high setpoint

low setpoint

soil moisture measurement

irrigation events



Ongoing Work: Experimental Setup



Human in the Loop

basestation irrigation

scheduler

John Lea-Cox Charles Bauers

soil moisture sensors at 12 locations

38% increase in #1 stems



Automatic Decision Making: Modeling Approach

ModelMapping to

Control Decision

Model Parameters

sensorinputs

modeloutputs control

signal



Example: Petunia Model [van Iersel et al.]

Model

Mapping to Control Decision (replace

amount of water used)

Model Parameters

varietyplant age

sensor inputs:temperatureRHlight

model output:water use irrigation

command



Feedforward Modification

Model(with

parameters)

replace difference

sensor inputs:temperatureRHlight

water useirrigationcommand

Model(with

parameters)

set daily irrigation schedule

weather forecast:temperatureRHlight

predictedwater use irrigation

schedule

At the beginning of each day:

At the end of each day:



Example: MAESTRA [e.g., Bauerle et al.]

Model



Model Parameters

tree location, geometry, s

oil type, LAI, leaf physiology…

sensor inputs:temperatureRHPARwind


command



Example: MAESTRA [e.g., Bauerle et al.]

Model



Model Parameterstree location,

geometry, soil type, LAI, leaf physiology…

sensor inputs:temperatureRHPARwind


command




G. KantorCMU Robotics Institute

Obrigado

• USDA SCRI CASC Project: CMU, Penn State, Washington State, Purdue, Oregon State, Vision Robotics

• USDA SCRI MINDS Project: U. Maryland, CMU, Georgia, Colorado State, Cornell, Decagon Devices, Antir Software

• Jim McFerson and WTFRC

• IHC 2010 OrganizersIHC 2010Lisboa25 August 2010


O Fim

Distributed Sensing in Horticultural Environments

Education