It is a very rare ProBot. It’s been around since 2019 and has gone on to consistently placed in the Top 8 of bots and has even taken several championships.
The wild part is that it hasn’t been updated since May 11th, 2022! which is nutty to think about.
What’s given it so much staying power?
There is an interview with MicroMachine’s makers, Raphrr and Unsual, that lays it out.
Let’s dig into it in this issue of README.
5 Rules of MicroMachine’s Staying Power
Economy First
If you watch a MicroMachine game, a keen observation is how aggressive it is. What many miss is how macro-heavy it plays, almost to the point of being too greedy.
Many of the games it wins are simply because it prioritizes keeping a healthy economy.
In the early days, it was really susceptible to all ins. The makers accepted those weaknesses and iteratively patched the obvious failure cases, all so they could have the resources to sustain aggression.
Do not hard-code for every matchup
One thing I see a lot of bot makers do is, when they encounter a loss, they create a special provision for that situation. Soon enough, they have a patchwork of edge cases and one-offs.
For MicroMachine’s makers, if something works on one map, it should work on all maps.
Tackle the fundamentals. Build systems that generalize. Fix your detection and core logic instead of writing twenty matchup-specific hacks.
Make decisions with data, not vibes
Every game is a treasure trove of data points, and MicroMachine’s makers take full advantage.
They track strategy win rates per opponent locally from the ladder server.
The bot then picks builds based on success rates. If one opener fails, it tries another, and statistics drive the choice.
Even very simple stats or “baby machine learning” can make your bot feel smart. Start logging results and let the bot adapt over time.
Micromachine’s cyclones tearing up opponents
Patches will break you, design for change
If there’s an event that all bot makers dread, it is patch day. The day when assumed mechanics can break, sending you into hours, if not days, of repair work.
For MicroMachine’s makers, the approach is a mix of API data and hard-coded values, and adjusting those when patches hit.
They keep all the game constants and weird timings in one place so patch days hurt less.
Open source makes you stronger
At one point in its history, MicroMachine was the strongest open source bot available.
This attracted others to test against it, find weaknesses, and even copy ideas.
It also let its makers see new counters and patch vulnerabilities faster.
It later did get exploited to the point that they decided to close it off, but I would still argue it would not have had this much staying power without that open phase.
Sharing your bot can feel risky, but it accelerates your own improvement and helps the whole scene.
These are some timeless principles I took away that you can apply to your own bot.
PiG_Bot
introduced a bug two weeks ago that left my bot helpless, both when attacking and when defending. I refactored and added squads so that different units can turn around and defend the base, but I forgot to put their responses in on_step, which led to them not being controlled properly 🤦🏾♂️
Improved attack micro for more persistent unit engagement and refined tank support logic. Units now rally to tanks when any nearby unit takes damage. Added banshee harass versus Protoss with refined cloaking, threat avoidance, and repair cycles.
Are you a member and want to showcase what you are working on? Shoot me a message.
The last ProBots of the year is coming up, and we have got the top eight bots from our qualifiers ready to showcase their work. Check out the Bracket lineup, it has:
Zozo
negativeZero
Eris
MicroMachine
Xena
BenBotBC
GPT
DominionDog
With DominionDog and GPT coming up from wildcards, we will see which bot can take the top spot this season.
