Game Design Desiderata
Written by Alistair Lynn, for SR2019.
The way I see it, there are 8 main requirements, or criteria to assess a game by:
- Game length,
- Proof against first-order optimal strategies,
- Reasonableness to prototype,
- Presenting axes of exploration,
- Practicality and robustness,
To go into more detail:
Scoring in games must always incentivize robot improvements. Specifically, making a robot more capable, intelligent and robust should produce a meaningful improvement in expected score. Games should also consistently rank better robots above worse robots.
There's quite a bit to unpack here. The key thing is that all levels of ability, making the robot 'better' should meaningfully improve your score. For teams right at the very bottom, going from a robot which does nothing to one which does something should improve their score. For teams with middling ability, moving parts of their strategy from dead reckoning to something more intelligent should improve their score. For the teams right at the top, they should be able to make marginal gains and clever strategy tweaks which will improve their score. This always encourages teams to push forward.
We want to avoid ties. A preponderance of ties is an indication that a game isn't differentiating well enough between some types of robots. This somewhat implies a wide range of ways to score points.
Modelling points scored as geometric (which is common for sports), we can arrive at the following table for how many points robots should be scoring to achieve some level of ties:
N. robots per game % of games with ties Mean score Median score 2 robots 50% ties 0.5 1 2 robots 20% ties 2.0 2 2 robots 10% ties 4.5 4 2 robots 5% ties 9.5 7 2 robots 1% ties 49.5 35 4 robots 50% ties 4.1 4 4 robots 20% ties 13.2 10 4 robots 10% ties 28.2 20 4 robots 5% ties 58.2 41 4 robots 1% ties 298.2 208
A corollary to this condition is that the game needs to have a low barrier to entry. In other words, from a cardboard box and some boards, we should be able to differentiate consistently between a team which assembles them in a basic fashion and a team which doesn't.
All games must take exactly the same amount of time. This is actually a differentiating factor from sports (which often have rules around tie-breaking, or pausing) and from something like Robot Wars (where games can end early). This is needed to be able to sensibly schedule the competition, since we are packing such a huge amount of matches into such a short space of time, but also means that robots experience an even amount of wear and tear.
Wear and tear is often a significant factor – robots are occasionally quite beaten up by the time we reach the finals.
Implied by the game length criterion and the Differentiation criterion is that robots should never be incentivised to 'camp' or 'body-block'. Games should be designed so that it's never in a robot's interests to, for instance, get some tokens into a zone and then switch off and no longer move to block other robots from getting in.
Proof against first-order optimal strategies
This is a bit of game design. A relevant blog post:
Essentially, there should be no one simple strategy which obviously dominates all other strategies. The game should be designed so that there are a range, preferably a very large range, of possible options. This improves spectacle, and makes the challenge more interesting, but it also avoids a situation where a small number of teams find this strategy and the competition devolves into separating the teams with that strategy and the teams without it.
Reasonableness to prototype
Teams need to be able to prototype their robots in schools. This means they need to be able to construct an environment which is a reasonable approximation of the one they will be facing at the competition. Some behaviours are harder to approximate and some are easier, but they key point is not to need complexity or spending to be able to understand how robots will behave in the field.
Game props and components need not be set up to be exactly replicated in schools, but should be easily approximated. Cardboard boxes and tin cans are very easy for schools to acquire, but if our setup involved a rope which robots needed to climb that would be difficult because it's hard to construct something stable enough to anchor that in a school.
Active components in the arena are a problem, but aren't entirely ruled out as long as it's possible for schools to approximate them (such as by having a person holding something and taking some action at the right time).
Presenting axes of exploration
Ideally games should present challenges in all the areas of robotics which teams can explore. Adding degrees of mechanical complexity where teams can explore building interesting actuators to gain an advantage are strongly encouraged, as are degrees of sensing complexity where teams can explore building interesting sensors, and tactical complexity where teams can explore building interesting software.
This must not, of course, increase the barrier to entry, but ideally the game should involve challenges on all three axes. It is also better if there are challenges on these three axes at different skill levels: that is, low-performing teams can still gain some advantage by exploring options at these three levels as well as high-performing teams.
Games must be fair. They must also be perceived to be fair, which is subtly different but also important. Aspects of this are making sure that the arena is symmetric so that no team is advantaged by their starting position, and avoiding randomness if it can be seen to have given some teams an advantage and disadvantaging other teams.
A good test of this: if a team described being "robbed" of an award by some aspect of the game, would you be able to keep a straight face? If one robot has easy access to some McGuffin which gives them lots of points and it's harder for the other three, one of the other teams could reasonably claim that they had a better robot and should have got an award but were "robbed" by their starting position. If, however, the team claimed to be "robbed" because they fell at a hurdle every other team had deliberately designed to be able to pass, we would be pretty scornful.
Practicality and robustness
Given that we need to actually run this competition, making the setup practical and robust is very important. Realistically there is limited volunteer time ahead of the competition to build things; a workable game should have a plan for buying and building everything needed for the arena, and should also plan for how we can reasonably reset it between matches.
It must also be robust, in the sense of minimising the chances of having to delay or cancel matches due to arena issues. Robustness can be simplifying components or building them to be physically stronger, it can also mean planning for having spares available, or using passive mechanisms instead of active ones.
Student Robotics is a spectator sport. It should be very entertaining to watch matches. This has a few implications.
The scoring system should be easy to understand and it should be easy to determine if someone is a long way out in front. Interaction between robots (non--destructive of course) should be strongly encouraged, because it's always fun to watch–and it also encourages intelligence. Robots should be constantly on the move and there should be plenty of opportunity for last-minute changes in who wins a match.
These are the desiderata which I recommend we use to think about game ideas. Obviously it's not everything, but I think this is most of what we're looking for from a good game.
Reference: https://studentrobotics.org/trac/wiki/Game%20Assessment%20Criteria, the original criteria used in SR years ago