owenbupt / ae-project Goto Github PK

About the notation:

- Q is the covariance for the noise in the motion model (like they use it in the
paper, contrary to the book).
- Rfeat is the covariance for the noise in the measurement model of the
features (when a robot measures a feature, as in lab1 and lab2).
- Rrobo is the covariance for the noise in the measurement model of the robots
(when a robot sees another robot).

Thoughts for the report:

- The update steps of the Kalman filter have been separated in observed and
observed (the robot saw another robot or was seen by another robot) so the
simulation represents better a real decentralized system.

- In order to handle the cross-correlation terms we have made the decision that
each of the robots takes care of one complete row of the centralized system
matrix. This means that the robot 1 for example would store information about
P11, P12 and P13. Robot 2 and 3 would store P21 and P23, respectively. This
might be a bit redundant since P12 = P21' and P13 = P31' but it is a somewhat
general representation.

TODO

- The non-initial update step for the robots that have not participated in the
measurement.

- Make in separate functions the computation of H_tilde, R_tilde