Hacker News Reader

Comment by anfractuosity | Hacker News Reader

Maybe the mice could use mmwave to possibly see through maze walls. Or use a little 360 degree camera on a pole. I assume this would be classed as cheating though.

Sounds a fun competition :) Looking at the Veritasium video they're crazy fast.

View on HN

Apparently, using a camera on a pole is not cheating, but it is not really necessary to "see through walls", as the robot has 10 minutes to score the best run, from start to finish, giving the robot enough time to explore the maze and figure out the best path. Only the fastest run counts.

The pole can make figuring out the best path easier, but at the cost of making the robot bigger and heavier, which I suppose is a problem when you see how fast the top contenders are moving.

Source: https://attend.ieee.org/r2sac-2020/wp-content/uploads/sites/...

At first, I thought this makes the problem of finding the optimal path rather easy, but on a second thought it's not, because the algorithm needs to take into account exact properties (max speed, acceleration/deceleration, turning), so each model of mouse may have dramatically different optimal routes.

I also imagine that mice will have a certain aggressiveness setting, perhaps even switching to a different route, using which you can achieve higher speed at the risk of a crash. It would be interesting to see if there are some strategies around it.

Thanks, I didn't realise that regarding the 10 minutes/fastest run.

This actually sort of ruins the sport for me. I'm sure there's practical reasons for it, just like with BattleBot but it's just so... lame. Put them into an irregular tunnel maze, time them from the millisecond they enter.

Maybe they are arbitrary but they are the rules. They are the reasons the sport is how it is today.

And considering that there is actual competition and innovation, these are good rules.

Feel free to organize your own competition using the "blind maze" rules, maybe it will become a success, but it will be a different sport, which will test different skills. They can coexist, not one lamer than the other.

But then luck becomes the dominant factor. Whichever robot decides to take a left turn at the right time will win.

And that's bad, how? Is it a sport (where the optimal amount of luck is non-zero) or a scientific study? Also just make them run multiple mazes then. I tend to think of sport fundamentally as simulated warfare, the further you go from the mess of reality in pursuit of fairness and justice, the less worthy the endeavor becomes.

Reducing luck is bad, how?

There are scales of it, reducing luck in the sense that you don't require a perfect course. That a speck of dust ruins your run etc. is, within reason, part of the game. But flying blind into a maze and just randomizing your way through it isn't particularly challenging. Just enter the game with a 100 identical robots and you'll have a better chance at winning.

Multiple mazes isn't a bad idea, but a massive endeavor just increasing the barrier for these competitions to take place.

>... the less worthy ...

Why? The most noble sport is a lottery then? Because that is essentially what you would turn this competition into.

No, I don't think random pathing is the best algorithm for maze exploration. Doing best possible with limited information in real environment should be the meat of this competition, not having all the time time to leisurely map the entire regular gridded space and then hyperoptimize a solution based on specific behaviour of tiny wheels on flat cardboard surfaces. Boston Dynamics robots can do backflips nowadays. As I said, I consider _some_ luck better than no luck or all luck. Entering with 100 identical robots... you'd have to actually build all of them, people would shun you and I'm sure nobody would be against a rule against that.

edit: But the accessibility argument is probably valid, yeah. There should be a higher less artificial league then, or something.

> Or use a little 360 degree camera on a pole.

that made me chuckle.

mmwave to see through walls at that size & speed can't be done yet, or can it?

I'm not sure to be honest. http://rfpose.csail.mit.edu/ is something I've come across before to detect pose of a human behind a wall.

Just found https://www.youtube.com/watch?v=IVpXeO0DYDQ - "Sense the world with TI mmWave: See through rain and walls, and detect small objects", which looks very interesting.

Ti seems to make a number of mmwave evaluation boards/chips.

It’s kind of redundant because they have a reconnaissance phase which is (relatively) untimed. So they can map out the maze at their leisure, then use that map for the time trial.

I guess if you can pull that off while still maintaining that speed, then it's not cheating. Those things are crazy fast!

Reply on news.ycombinator.com