Subject:
|
Kalman Filter on the NXT
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Fri, 28 Dec 2012 12:58:05 GMT
|
Viewed:
|
20285 times
|
| |
|
|
The Kalman Filter may be perceived as a discouraging, almost occult calculation
method. Many do apply the filter just by following some "cooking recipes", and
they don't care much about the internals. However, if the filter is well studied
and adapted to the current problem, then --tada-- things may appear very easy...
and its use on the NXT is no secret anymore.
If you want to learn more about the Kalman Filter, then have a look at its
implementation into a simple application on www.convict.lu/blog/?p=230. A
vehicle drives don't the inclined plane, and wants to exactly know its speed and
position. It is equipped with an accelerometer, a tachometer -made of a LEGO
legacy motor, used as a DC-generator- and a legacy rotation sensor. Every sensor
has its own trouble of accuracy and precission:
- the accelerometer is caracterized by an important drift, making computation of
the speed, and position (by integration) very error-prone
- the tachometer has a significant ripple due to the motor coils.
- the rotation sensor has a lack of precision, because of its bad resolution
(1/16th)
Now, the Kalman Filter overcomes all of these issues, by considering the system
model, which allows to predict, how the system will behave. Then the
measurements are used to correct the predictions. The predictions are based on
Galilei's equation of the motion. However, the system will not exactly obey this
equation, because the world introduces errors: the plane is not totally even,
for instance. The Kalman Filter merges everything, the sensor readings and the
model estimations into the best possible estimation of the system at any time.
|
|
Message has 1 Reply:
4 Messages in This Thread:
  
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
Active threads in Robotics
|
|
|
|
|
|
