mirrcub3r | The (formerly) fastest Lego Rubik's Cube solving robot is now even faster (1.2s avg.)!

As of February 2020, the world's fastest Lego cube solver is my very own SquidCuber (https://youtu.be/wLzn1w8vgM4) with an average of 1 second flat!

If you have not seen the initial version of mirrcub3r, please check it out first: https://youtu.be/03V3915YuB4. The video / its description will probably answer all the most common questions about the mirrcub3r design in general. Here, I will solely focus on the improvements that were made since then. The full source code is again available here: https://github.com/efrantar/mirrcub3r (robot), https://github.com/efrantar/rob-twophase (solving algorithm).

The initial version of mirrcub3r published about a month ago was already pretty fast, but it still had one rather big "flaw". Concretely, 2 of the 5 axes were only driven by a single motor (and were thus considerably slower) as 2 Mindstorm bricks only offer 8 motor ports in total. I have now added a 3rd Mindstorm to address this. Furthermore, now that all 5 sides perform the same, I was able to reintroduce my corner cutting optimization system (which was just far too tedious and unreliable to tune for the prior version), saving another ~50-100 milliseconds on average. Basically, the robot distinguishes about 20 different cases of consecutive turns (ex. clockwise into counter-clockwise) with specific tuning for each of them. This is useful because some of those situations run much smoother than others (due to the mechanics of the physical cube) and thus permit much more aggressive move scheduling. Additionally, the control program also figures out the optimal directions for executing 180-degree turns (which can obviously performed both clock- and counter-clockwise) with respect to this corner-cutting. With some great tips from Mike Thomas plus some ideas of my own I was also able to improve the robustness of the color matching considerably (although this has no effect on the solving speed).

While the average of this version is around 1.2 seconds, with a bit of luck it is actually feasible to get Sub 1 solves on real scrambles making mirrcub3r now the first ever Lego robot to solve a Rubik's Cube in under 1 second. In particular, the 0.929 shown in the video is a 17 axial quarter-turn solution, which is certainly quite lucky but definitively not "winning the lottery" lucky (as the bonus-solve in my previous video with 12 axial quarter-turns was). I got this time, after manually trying out about 30 fully random scrambles. Depending on the particular corner cutting situations occurring in the solve (and some luck regarding how the robot shakes throughout the solve), Sub 1s with 18, 19 or even 20 should potentially also be possible.

Now you might ask: Is this it, can one go any faster with Lego? Well, it is not over, at least not quite yet. While I am very happy that I managed to reach the Sub 1 milestone already with mirrcub3r, I will certainly still be going for a 6-axis construction (meaning shorter solutions due to be able to turn all 6 faces) next (this will unfortunately mean having to drop the mirrors). We will see how close Lego can get to a 1.0s average. :)

Видео mirrcub3r | The (formerly) fastest Lego Rubik's Cube solving robot is now even faster (1.2s avg.)! канала Elias Frantar