Adaptive sampling in environmental Robotics

1
Adaptive sampling in environmental Robotics

• Mohammad Rahimi, Gaurav sukhatme, William Kaiser,
• Mani Srivastava, Deborah Estrin

2
Motivation
NIMS Networked Infomechanical Systems
3
47 m
Solar Cell
91
NIMS Node
84
Visible Imager With Pan/Tilt Actuator
Power Distribution Cable
50 m
Met Node (Ta, RH, PAR)
Battery Pack
Data Base
Telephone/ISP
Internet
WRCCRF Crane Site Dry Shack
4
Science Objectives
H2O
Q
C
C13/C12
T, RH, Wind
T, RH, Wind
C13/C12
T, RH, Wind
C13/C12
T, RH, Wind
Growth
Growth
5
NIMS Prototype Deployment
6
NIMS Prototype Deployment
7
Problem

• Creating a dynamic Map of the environment
• Based on the carrying sensors (attributes)

8
Approach

• The robot (shuttle) is an agent
• gather Geostatistics information
• Refresh those statistics as fast as possible

9
Digitizing robots world
0,0
Shuttle Patrol Area
Obstacles
Cell size that we call it a pixel is a xx.
pixel is the distance that shuttle moves
atomically
10
Assumptions

• Shuttle is certain about the location
• Sensor reading error is zero
• Environment is static in circuit convergence time
• Warning to the user to reduce coverage or
  expected accuracy otherwise

11
Sampling Policy

• Stratified Sampling
• Divide the population into subpopulations
• Extremely better performance with some degree of
  apriority domain knowledge
• Random sampling
• Mean proportional to cell size

12
Feedback

• Using variance of data to classify a region
• Vaiance/Mean lt Expected error
• or
• Variance lt Sensor Noise

13
Divide and Conquer

• Stratify the current cell into four
• µ a cell size (µ is mean of step size)
• Collect data in current cells (Random)
• Calculate the variance
• Iterate until variance is below threshold

14
Closed Loop System
Acceptable error
Estimation
Stratification Policy
Map
Reading points

-
error
15
Result Of the Algorithm

• Quad-tree Map of the variance of the environment
• Shuttle step-size is random but proportional
• to how deep in the tree it is

n
Log (n)
n
16
Initial Results
17
Wish List

• Adding time domain
• Static sensors as sample support