Sunteți pe pagina 1din 2

Mobile Robots Homework assignment

Kalman Filter with accelerometers


Friday 23rd December, 2016

In the lecture we discussed the Kalman Filter. Your individual theoretical homework assignment
this week is to apply the Kalman Filter on paper to a very simple robot. Consider a robot which is
equipped with a synchro-drive and two accelerometers. The accelerometers are mounted such that
their main axes of measurement are perpendicular to each other, and that the first accelerometer
measures in the direction of the x-axis of the robots own reference frame. The robot has no other
odometry sensors which could provide additional information about its traveled distance or speed.
It has some sensors to detect landmarks for using in the measuremend update steps of the Kalman
filter, but this assignment only deals with the motion estimation part of the Kalman filter.

Tasks
1. Derive the motion model, i.e. give equations to calculate each of sx , sy , vx , and vy for the
time t based on the measurement at the time t and possibly their previous values in state
t 1.

2. Instantiate the matrices A and B in the Kalman filter algorithm based on your calculations
in the previous task.
3. Assume each accelerometer has a Gaussian error distribution with an average error range of
= 2% of the true value, i.e. for a true value the probability to measure x is given by

1 (x )2
p (x) = e
2 2

with = 0.02.
(a) Give the covariance matrix u for the two sensors uncertainty.
(b) Calculate R as a function of u and the other ingredient of R mentioned in the class.
4. Optional task, +3 extra points. Assume the robot has a docking station which period-
ically sends out a guidance signal. The robot can receive this signal only when it is located
within 4 meters from the docking station. The robot has one additional sensor, which within
the limits specified before can detect the direction towards the docking station, bot not the

1
distance. It has an absolute error of 2 , independent from the sensed value. The docking
station is known to be at the origin of the global reference frame, and the robots reference
frame is initially aligned with the global reference frame.
(a) Can this additional sensor be described by a linear error model as in zpredict = C?
Provide a brief explanation to your answer.
(b) Instantiate the sensor model, either by giving the matrix C or a nonlinear function
involving (possibly together with some other ingredients) the state .
(c) Instantiate the sensor error model (4th Kalman equation), either by giving the individual
matrices involved in the linear model (including Q), or by giving needed Jacobians in
the non-linear case. If the non-linear case applies, it is sufficient to write the elements of
the Jacobian matrix in the form of variable

f unction (arguments).

Hints
From physics we know s = vt, v = at, and s = 12 at2 , where s is the distance traveled, v is the
velocity, a is the acceleration, and t is the time passed.
The robots state, split into x/y coordinate components, is described by the state vector
[sx , sy , vx , vy ].

The measurement update is given by the two accelerometers readings [ax , ay ].


Here u and in consequence R are dependent on the actual value of the update measurement
u at each time step.
If you dont remember what else, apart from u , goes into R, have a close look at the second
Kalman equation and ask yourself what dimension R must have and what is actually done in
the first part of that equation.
For task 4(c), you are on the right track if S is of a very low dimension. (Please note that the
numerical calculation of S is not required!)

This assignment has to be solved individually, no group work allowed.

Please report the time spent to solve this assignment.

Due date: January 5th

Good luck!

S-ar putea să vă placă și