[robocup-humanoid] Roadmap Feedback

Josiah Walker josiah.walker at uon.edu.au
Sun Feb 2 19:33:34 EST 2014


Hi TC (and others),

Speaking for the NUbots, in general the roadmap presented is a very good thing. We believe a stable roadmap with major rules-changes every 5 years is a great plan, as it allows us to plan ahead and reduce our total cost/workload for the competition while focusing more on research. We should also see teams getting more advanced and competitive toward the end of each 5 year period, so long as the rules don't change significantly during those periods. This allows the public to see more "exciting" games. There are several decisions we wonder about however, as they seem to be working against the interest of advancing robots towards playing with humans.

We have observed in the past two years of being in the Kidsize league, that with the exception of 1-2 top teams the league's game-play and strategy is little more than "run to the ball and kick in a pre-defined direction". As long as only the mechanics of the field and team size change, and the core rules do not, then this may be as advanced and exciting as the league will get. I know there was a significant amount of discussion about the direction changes should take at RC2013, but would ask the TC to consider rules which challenge teams to overcome the following specific situations which we observed:

- Current hardware rules allow the construction of robots which can stand in one place and turn only their neck to view the whole field. This encourages the strategy of standing still and scanning without repositioning as more accurate measurements can be obtained (especially for larger robots, less mobile). Such a strategy is effective only because the hardware requirements on neck swivel and field of view make it reasonably easy to implement, but it makes the gameplay less exciting and allows robots to get away with the "easy route" in terms of vision and reduces the incentive to overcome error induced from motion.

- As far as we read in the rules, magnetometer type sensors are still allowed. As of now the NUbots run on a robust multi-hypothesis Kalman filter localisation system which we can extend in the future to add more dynamic features. However if these types of sensors will be allowed for the next 3-4 years, then it is in our interest to use these sensors and remove much of our vision based localisation code. This is especially true as robots increase in size and there is opportunity to place the sensors further from motors in order to obtain better readings. This encourages stagnation in the vision development of the league.

- Wireless communication has always been a sore point at Robocup. In fact it was so bad at 2013 that in the majority of our games, teams were not properly penalised for ignoring the game controller. Alternate methods of communication would bring us one step closer to human team levels of communication and coordination.  RGB LED signals (as we are abandoning colour on the field) or sound/gesture based communication could be used. Requiring teams to intelligently process the extra sensory information this entails, and to be smart in the choice of data to communicate drives innovation and research at the same time as giving something extra (or more "radical") for audiences to appreciate.

- Robot detection and avoidance is almost non-existent. This often causes falls, damage to robots, and slows the game down. There are currently rules for this, but they are ignored in practice. Having worked on this problem for a while, we acknowledge that it is very hard, but also that it is very important to develop for playing against humans. Additionally, in the majority of gameplay cases, actively trying to avoid opponents will put the conforming team at a greater disadvantage (we disabled our avoidance at less than 50cm due to this). I am not sure what the best solution is here, but would ask the TC to consider options which advantage teams who try to conform. Possibly penalty kicks rather than robot removal (if this doesn't slow the game down too much).

There are many more examples of these simplest-wins strategies, which we are sure the TC has also observed in games (we may even be guilty of some - such as removing robots from the field who cannot communicate wirelessly to avoid teammates). Some of these concerns are addressed much later in the proposed roadmap (such as the need for robust localisation) however delaying the requirement for these less expensive, more intelligent changes will likely see a league which focuses more on speed of getting to the ball and force of kicks, and less on intelligence and interaction during games. We do not think it will be sufficient (or even safe) to have these types of robots playing human players in the future, so we would ask the TC to consider delaying the merger of adult-size and teen/kid-size into one league by 5 years or so, and instead focus on requiring the teen/kid-size league to comply with more realistic/intelligent rules at a faster pace. This allows us to have a (more "radical") demonstration of what is possible progressing faster than the hardware progress of the larger robots.

Regards,
Josiah (and the NUbots).?



More information about the robocup-humanoid mailing list