[robocup-humanoid] Significant concerns regarding the proposed humanoid league roadmap

[Marcell Missura]

Dear Sean,

thank you for voicing your concerns. We all can profit from individual opinions brought forward and giving rise to discussion about the future development of the league. Please allow me to address your concerns.

> >From this vantage point, I think the most important reasons for RoboCup are:
> >advancement of robotics research, promotion of robotics teaching in
> >universities, and promotion of robotics in the worldwide public eye.  One
> >reason I think is *not* an important goal is to produce a robot soccer team
> >which will defeat the World Cup team by 2050. 

The challenge to defeat the human world champion in the middle of the 21st century is what the RoboCup initiative was constituted upon (http://www.robocup.org/about-robocup/objective/). Having this clearly defined goal gives us the necessary focus and motivation to direct our research towards achieving a technological impact with undoubted practical implications. In my personal opinion, having this goal will only promote advancement of robotics research and teaching, and raise public awareness, for example when robots start playing against humans. Without this goal, we are researching aimlessly at best with the motivation to defeat the currently leading team, but without the drive to invent useful technology with a practical application. 


> 1. I would like to know if other leagues are being asked to produce
> similar documents which outline how they will be continually physically
> modified until they produce robots capable of defeating the world cup team
> by 2050.  If not, are these leagues expected to be phased out?

We have not been asked to produce this document. It emerged based on discussions and ideas in the most recent years and by the "nudge" the trustees gave us to try and make more radical changes. Having a long term plan all the way to 2050 gives us a rough idea of how large our steps need to be in order to reach our goal, and to keep an eye on the oncoming changes so that they do not take anyone by surprise. This is useful, no matter what other leagues do. I do not know the future of the other leagues. In the simulation league for example, no physical robots are used, so I don't think they will ever be expected to "produce" one. Every league has the purpose to focus on solving a task with reduced difficulty. All leagues are encouraged to borrow insights and to copy technologies from other leagues. I believe the Humanoid League and now the SPL are the potential candidates where the best results of all other leagues could be merged and the 2050 soccer team can emerge from.


> 3. The proposed changes are extraordinarily expensive. [...] If we
> want to use humanoids as our platform, our only option will be to form
> consortia; and those who cannot will be left out. 

The roadmap barely even touches physical changes to robots. The focus of the roadmap is the change of the environment and the encouragement of certain skills needed to play soccer at human skill level. Concerning hardware, we try to be as little restrictive as possible. Our intention is to encourage the construction of larger robots in larger numbers. And larger robots in larger numbers cost a lot more money, this is certainly true. But, be it for achieving the 2050 goal, or for learning how to build and control robots with an adequate size to operate in human environments, small sized and cheap robots will not be a useful research tool for much longer. We must make the effort and find ways to support more realistic soccer players. That said, the roadmap has been laid out in a way that no one will be excluded and no one will be required to spend more money, than they already do. In fact, the first potentially cost effective change the current roadmap document foresees happens in 2020, when the minimum height for qualification will be raised from 40 to 60 cm. It is not unlikely that by that time robots of this size will be available for the same price as stock KidSize robots today. Furthermore, the roadmap suggests somewhat radical changes every five years, but with the bonus that robot specifications and the environment the robots play in stay constant for five years. This serves as a protection of your investment. Now, to be able to play soccer with larger and more expensive robots, the solution will potentially be to form mixed teams. One university can bring only one or two robots that will be joined with others to form a soccer team. One or two large robots are not more expensive than four of five small robots. The idea of mixed teams has not been addressed by the roadmap so far, but it has been proposed several times in the team leader meetings. We will have to work out the details of the implementation in the near future in a way that no one will be left out.


> 2. The proposed document is heavily aimed towards the mechanical and control
> aspects and capabilities of the robots, and does not touch significantly on
> the enormously problematic computational side of things.  Given the makeup
> of the non-SPL humanoid leagues this is an understandable but large
> oversight.  It is true that soccer presents a simpler computational
> environment than many other real-world scenarios, but even so, I see few
> milestones or technical challenges in the document which directly address
> the vision, multagent coordination, learning and adapting, planning,
> probabilistic reasoning, or speech recognition tasks necessary to create a
> viable team by 2050.

It has never been our intention to define how certain aspects of the task will be solved. The roadmap can only suggest the necessary environmental changes and useful skill checks up to 2050. Whether the best soccer robots will use learning, planning, or reasoning, is entirely up to future research. Roughly half way through the roadmap (2025) you find yourself on a 30 meters long FIFA-like soccer field with white lines, white goals, most likely with a size 5 white-ish ball, and a grassy surface, where fully autonomous games are played including referee interaction. This is where you will have to be able to see and localize, to walk/run, to coordinate with your teammates, and to follow the instructions of the referee. The environment itself and the task within is the challenge for computer vision, AI, and motion generation. 


best regards,
Marcell Missura
member of the Technical Committee