Currently odometry only takes the last motion commands and assumes that the motion succeeded and takes it as odometry. In Hannover we have seen that this approach is likely to fail. On the white (tape) lines there was more slippage than movement, when the walk parameters were not optimal (and we have not seen a situation, yet, where they were optimal, and probably never will with ALMotion) the robot usually diverges from the intended path. Remembering our last game in Hannover when walking "straight" we often ended up walking on a 90° arc, caused by tape and bad walk parameters.

Experiments are to be conducted to base the odometry on real sensor values, likely on the intertial measurement unit (IMU). Work has been done in Graz to record ground truth data to come up with a mapping of sensor readings to expected actual movements. This needs to be integrated. A good odometry is required for good position tracking in the localization in the time when no lines are seen, for instance if the head is currently focusing the ball.