Files
Abstract
Gathering information on robot motion is critical in determining how a mobile robothas interacted in its environment. The information about a robot’s motion, called
odometry is generally collected by tracking wheel behavior for ground based robots.
Wheel encoders are a typical means for collecting this information. Wheel encoders
generally split the circumference of the wheel into equally discrete distances to track
the distance traveled along the wheel’s circumference. The proposal of this research is
to dismiss the use of the above discrete based wheel encoders and use an accelerometer
as a wheel encoder for a more continuous reading of the wheel’s position.
Accelerometers are typically difficult to use for precise data collection because of
noisy outputs causing inaccurate odometry information. This can be overcome by
building a model to predict how the output of the accelerometer on the wheel should
behave. With the accelerometer mounted on the wheel of a ground based robot, an
expected output of the accelerometer should oscillate between positive and negative
values of the gravity vector. This allows for a system that can be modelled as a
sinusoid.
Using the sinusoidal model, the raw data can then be filtered with an Extended
Kalman Filter (EKF). The Extended Kalman Filter weights the measurement from
the sensor along with the predicted value from the model to give a low noise, high
precision output. The results from the newly filtered data proved to give acceptable
accuracy that was desired for estimating position of the wheel. Based on the filtered
data, the calculated distance proved to remain within %2.2 of the expected distance
traveled for each trial run of the experimental. The trend of the error also leads to
the belief that this error is not likely to increase inside the testing environment.