Localization

1
Localization

• Life in the Atacama 2004Science Technology
  WorkshopJanuary 6-7, 2005
• Daniel Villa Carnegie Mellon
• Matthew Deans QSS/NASA Ames

2
Basic Description

• Sensing
• INS 3 axis accel, 3 axis gyro
• Sun sensor
• Encoders
• Inclinometer
• FOG
• Motion commands
• Estimation
• Kalman Filter
• Nonlinear smoothing
• Dedicated PC-104 stack.
• Goals
• Accuracy 5 distance traveled
• Orientation within 3
• Odometry within 2 of distance traveled

3
Block Diagram
4
Sun Sensor
Now includes integrated inclinometer
5
Sun Sensor
6
Sun Sensor
Camera model error 0.5 pixel RMS 0.250 RMS for 3
dimensions of rotation Integration Problems
with h/w and s/w integration In the field A few
degrees Obvious systematic errors calibration?
7
Dead Reckon Estimator

• Straightforward path integration
• Relied only on data from sensors
• Encoders
• FOG
• Roll-Pitch
• Does not use
• IMU
• Sun tracker

8
Dead Reckon Results
9
Kalman Filter
system outputs (sensors)
rover
system inputs (speed, radius)
_
predicted outputs (sensors)
model
predicted state (x, y, z, roll, pitch, yaw)
K
updated state

10
Kalman Results

• NaN

11
Nonlinear Smoothing

• Performed when robot is stationary
• Operates on a sub-sampled sensor dataset
• Revises movement history
• New pose and covariance fed back into filter

12
Nonlinear Smoothing Results
Simulation
Filtering
Smoothing
Heading correction propagates to corrected
position
13
Next Steps

• Sun Sensor
• Early specification of interfaces
• Better coordination of efforts
• Estimator work
• Kalman Filter debugging, improvements
• Comparison of Kalman vs dead reckon
• Real-time Linux kernel

14
Next Steps Visual Odometry
Accuracy of 0.1 1mm over 1m 1cm at
5-10m Critical element of single cycle
instrument placement Could enable some return to
site/point capability
15
Next Steps Visual Odometry
16

• end