For a useful game play we need a clue where we are on the field. The bare minimum is knowing the orientation on the field to be able to turn to the opposite goal when in ball possession. This can be achieved by only looking at the goal colors (cf. ticket #85).

However, a full localization is still required. A few things that are not possible without full localization:

• Decide if we can kick reliably towards the opponent goal. When too far we might not even make it to the goal, or the ball is unlikely to go on a straight line for long enough
• Walk to certain "ball search points" on the field if no robot sees the ball
• Communicate ball position to other robots
• Position the defender appropriately
• Goalie cannot stay reliably in the goal
• No autonomous placement before kick-off, leading to a penalty of 0.5 meters towards the ball compared to other teams doing autonomous placement

A first base for the localization has been written. It now needs to be extended, adapted, or tuned to actually work during the game. We can happily take a localization that gives us an estimate only a few times a second (given the current slow game speeds). It just needs to be reliable enough to work over time, probably in combination with odometry (see ticket #86). Goal position data can improve the situation and is required to solve the symmetry problem (see ticket #85).

In the longer term the current approach must be re-evaluated, as it might prove not a choice as good for the Nao as originally expected. It stems from the MSL, where you have a higher chance of seeing lines due to the omni-vision, making it suitable for scan matching. For the directed Nao camera this is probably not the case, and head movement is a shared resource with the ball detection. The B-Humans and Nao Devils Teams are using a Monte Carlo-based approach, which seems a reasonable candidate for further investigation (after Graz).