[nathanww]

If you're really worried about accuracy, a three-stage hybrid system might be useful

1.The robot tracks the ball using the camera
2.When the ball becomes too close to track, or if the camera loses it, the robot moves into dead-reckoning mode
3.When a broken-beam sensor is tripped, the robot stops, kicks, and can start the cycle over again if neccesary.

That said, the balls don't move that much after they're placed. Dead reckoning is very susceptible to random variations in hardware, but a good encoder or inertial nav system can fix this.