Matt Kucia

Cloud, software, hardware

/projects/gy80

Sensor fusion

Overview

The aim of this project is to learn sensor fusion algorithms and implement them on ARM microcontroller.

GY-80

GY-80 is a cheap sensor board. Available on Ebay and DealExtreme.

http://dx.com/p/gy-80-bmp085-9-axis-magnetic-acceleration-gyroscope-module-for-arduino-145912

Board feature 4 sensors providing in total 10-dimensional information.

Sensor Description I2C Address (8bit)
L3G4200D ST three-axis digital output gyroscope 0x69
ADXL345 Analog.com 3axis Digital Accelerometer 0x53
HMC5883L Honeywell Three-Axis Digital Compass 0x1E
BMP085 Bosh Digital Pressure Sensor 0x77

Project assumptions

The purpose of this project is to provide sensor fusion solution using low-cost sensor board.

Phase I

  • Build a I2C ↔ PClink
  • Set up sensors
  • Read data periodically
  • Plot the data

alt

Launchpad connections

  • Serial transmission parameters: 115200 8N1
1.1 GY-80 VCC_3.3V
1.10 GY-80 SDA
1.9 GY-80 SCL
GND GY-80 GND

Source code

Screenshots

alt

Notes:

  • Scaling is being adjusted in real time, shake sensor board to set maximum values so graph can be scaled to fit the window.
  • Keys 1 to - turn on/off plotting value
  • Key c clears the window

Phase II

  • Implement 1D Kalman filter

Screenshots

alt

Notes:

  • Key k - toggles filtered graph
  • Key m - toggles measured value

Source code

Phase III

  • 2D Sensor fusion

Following graph show angle measurement using accelerometer (red) and gyroscope (blue). Gyro clearly shows error-induced drift.

alt

Complementary filter

alt

Complementary filter is a good alternative for small systems. http://web.mit.edu/scolton/www/filter.pdf

Comparison

alt

Filters:

  • Red - Complementary
  • Green - Kalman

While both methods provided unbiased value, the Kalman filter provided more stable readout. Present-day MCUs provide sufficient power to use Kalman filter in real-time.

Phase IV

  • Build balancing robot.

http://www.youtube.com/v/CwFBXYt4UKg?.swf?420×315

The robot consists of 5 parts: Tamiya gearbox, double H-bridge driver, Bluetooth wireless module, Stellaris launchpad board and GY-80 sensor board. The power is provided externally.

GY-80 board provides accelerometer and gyro sensor measurement at 100 [Hz] (UPS variable). Sensor data is then processed by kalman filter and feed into PI controller. Control signal is driving PWM output driving motors H bridges.

Notes:

  • The stability is good but not perfect
  • Kalman filter response was rally bad. I have boosted the response by multiplying angular acceleration value. Estimated angle value has overshot now but is fast enough.
  • Robot motors are powered externally. Wires are influencing robot stability.
  • Taller robot would be much better (bigger moment of inertia).
  • Both robot wheels are independent. Connected wheels would work much better reducing yaw.

Source code

alt

Robot operation. Red - accelerometer angle. Green - Gyro angular acceleration. Blue - estimated angle. Yellow - PI control signal.

References