An interesting smart cube idea I thought of

[Caleb Valenzuela]

A few disclaimers: I am not sure if this belong here or the “General Speedcubing Discussion” so mods please move it to the correct area if it is in the wrong place. Second, I am not planning to make this smart cube idea so i have no idea if it is even possible. So you critics say whatever you want, i just wanted to share this just in case it is possible and so that we can possibly implement it in the smart cube industry.

So i was thinking of chess.com one time and thought of their game review. I then thought of reconstructions and wondered, “could we race against bots in speedcubing?”. It seemed like a cool idea so i thought about it a bit more. Here’s the full concept:
Bots and Smart cubes
So if you take a look at chess.com, you can see that there are bots you could play against. In my concept here, you can race against bots using your smart cube (the bot uses a virtual cube). It could have a difficulty level like elo in chess. The harder the bot becomes, the more robotic and efficient the solutions are. The “bots” would only be solving a virtual cube while the user would be solving a smart cube.
Racing against “ghosts” of cubers
Users of the smart cube app can submit reconstructions of their own solves or others’ solves using reconstruction websites like reco.nz or twizzle. Alternatively, the developer could submit the reconstructions himself/herself or have the app track users’ solves and input it in the database. So let’s say the website/app has a “Yiheng bot”, which means the bot will try to accurately replicate Yiheng’s solving style using the database inputted from other users or the developer. Of course, there are some drawbacks such as a lack of human errors such as lockups and mistakes.
The app could also record your solves, which means it would also form a database of your solves so that you could race against yourself!
Solve Review
Like the game review in Chess.com, the computer would also review your solves showing the best moves you could have done and your “blunders”. This would help cubers see their mistakes and hopefully improve!

I feel that these features would make smart cubes more fun and make feel more worth it to buy a smart cube rather than just being able to see your splits and race against other cubers.

 

[pTr]

Future visionary.

 

[DA_BOY]

that sounds cool, it would be nice if they added the ghost part to smart cubes

 

[Deleted member 36445]

speedcubeDB has a type of "bot" that creates optimal solutions utilizing CFOP.

3x3 Scramble Analyzer - Speed Cube Database

3x3 Scramble Analyzer - Speed Cube Database

www.speedcubedb.com

battling an AI with real cubers solving style is an interesting idea. I think we have enough data from videos and reconstructions to make something like that at least half accurate right now using LLMs (even without smart cubes data)

if all the work is viable and has an substantial return it could be made some day.

 

[Kivvi]

This concept is really interesting. To simulate something like lockups, it’s probably possible to code in a chance for a short delay during some turns. The easier the bot, the higher the likelihood of such a lockup occurring.

Submitting solve recognitions to train an ai on one specific solving style is quite difficult I think, I’m not sure if that would work out that well. Instead, just programming the LL algorithms a bot uses might be more efficient. That way, you could have a bot that knows full ZBLL (Tymon/Xuanyi), one with 2LLL and fast recognition (Yiheng), one that takes forever to recognize his OLLs (me), and so on.

 

[Caleb Valenzuela]

speedcubeDB has a type of "bot" that creates optimal solutions utilizing CFOP.

3x3 Scramble Analyzer - Speed Cube Database

3x3 Scramble Analyzer - Speed Cube Database

www.speedcubedb.com

battling an AI with real cubers solving style is an interesting idea. I think we have enough data from videos and reconstructions to make something like that at least half accurate right now using LLMs (even without smart cubes data)

if all the work is viable and has an substantial return it could be made some day.

interesting but the main intention of my idea was to bring more reason to buy a smart cube because in my opinion, right noe there is not really a reason to buy a smart cube. Sure, you could see your splits with it and race with other cubers, but aside from that theres nothing else really worth mentioning.

This concept is really interesting. To simulate something like lockups, it’s probably possible to code in a chance for a short delay during some turns. The easier the bot, the higher the likelihood of such a lockup occurring.

Submitting solve recognitions to train an ai on one specific solving style is quite difficult I think, I’m not sure if that would work out that well. Instead, just programming the LL algorithms a bot uses might be more efficient. That way, you could have a bot that knows full ZBLL (Tymon/Xuanyi), one with 2LLL and fast recognition (Yiheng), one that takes forever to recognize his OLLs (me), and so on.

I mean if chess.com could recreate Magnus Carlsen and Hikaru Nakamura’s playing style then why can’t us cubers have that too? Cuber’s solving styles are relatively easy to distinguish for higher level solvers. For example, Yiheng likes to influence F2L as much as he can during a solve so the computer would find a solution that would influence as many pairs as possible while trying to be efficient. Tymon likes to go for those crazy pseudo XXX-crosses and do ZBLL so the computer would try replicate that.
That’s the reason why i think reconstructions would be necessary to pull this off. Without it, the computer would probably not be able to accurately replicate a cuber’s solving style.

Also could someone knowledgable in this field analyze this, thanks.

 

[Deleted member 36445]

interesting but the main intention of my idea was to bring more reason to buy a smart cube because in my opinion, right noe there is not really a reason to buy a smart cube. Sure, you could see your splits with it and race with other cubers, but aside from that theres nothing else really worth mentioning.

If you don't see a reason to buy a smart cube, then don't buy it

I use it mainly to analyse my solves and to drill subsets. I even practice megaminx last layer with it.

Megaminx LL Trainer with Bluetooth Support

 

[Caleb Valenzuela]

If you don't see a reason to buy a smart cube, then don't buy it

I use it mainly to analyse my solves and to drill subsets. I even practice megaminx last layer with it.

Megaminx LL Trainer with Bluetooth Support

I mean it would be cool to race bots

 

[abunickabhi]

Sounds cool yo

 

[ih8bld]

I use it mainly to analyse my solves and to drill subsets. I even practice megaminx last layer with it.

How do you do megaminx algs on a smart 3x3?

 

[CR7GoatSiuuu]

To make Ghost Bots more realistic you could analyze 100s of solves to get their exact recognition/execution times for their algorithms, as well as their move count

 

[Deleted member 36445]

@Caleb Valenzuela what's funny?

How do you do megaminx algs on a smart 3x3?

 

[Caleb Valenzuela]

If you don't see a reason to buy a smart cube, then don't buy it

I use it mainly to analyse my solves and to drill subsets. I even practice megaminx last layer with it.

Megaminx LL Trainer with Bluetooth Support

because this says you use smart cubes to train megaminx algs when there is only a 3x3 smart cube

 

[Caleb Valenzuela]

To make Ghost Bots more realistic you could analyze 100s of solves to get their exact recognition/execution times for their algorithms, as well as their move count

that’s what i was thinking too, which is why there is a need for recons. But i didn’t think of recog and exec times!

 

[CR7GoatSiuuu]

that’s what i was thinking too, which is why there is a need for recons. But i didn’t think of recog and exec times!

I was thinking Rubik's make it, all smart cubes connect, they have sensors in core to film ur hands as well, they connect cstimer to desktop and the stackmat mobile app to the mobile, have the rubik's games, maybe the rubik's slide game on the mobile and have like an ai estimated time for a scramble: after analyzing ur solves, it shows what it thinks u would have done; ur predicted solution and tps and compares ur actual time to it's prediction to gauge the strength of ur solve