Its upside down because its mounted to the roof, the camera probably has a mount on the bottom, so they just mount in inverted and flip the video later. This is also common practice with projectors.
I thought the first go where it was learning the maze (I believe) was the actual run and I was very confused. Then he zipped through the fuckin maze like a champ and it made me happy.
@applemauzel yeah, there was one I watched that was the winner for that year (I want to say it was 2015 but I am not sure) That the first 3 runs were pitifully slow and constantly going in circles/dead ends. Then the 4th and 5th run was a breakneck blur to the end and back. You could hear the jaws dropping in awe. 😆
@@Zhengrui0 We might be worse at memorizing mazes, but I'd like to see that little robot mouse design a little robot mouse that can memorize mazes.......
Anyone else marvelling at the code/automation script written for this and the amount of corner scenarios they'd have covered to tackle every possible situation ❤️
@@owenedwards8514 grip and precision are 9/10 the mechanical requirements... then a microprocessor and several distance sensors are going to have to fit into the micromouse... Theres a fair bit more that goes into it than "just buy an rc car". That only gets you as far as having a chassis, power supply, motors, wheels (they'll be shitty slippy plastic wheels and you are going to have to replace them) and you'll have to rewire the motor/motors to be controlled by an onboard processor instead of radio... It'd make more sense to start from scratch, imo.
@@CarlosFlores-nl4gg Nop. It would only be mean if I called him stupid. There are 3 possibilities: 1. Really smart 2. Mid wits 3. Stupids Since most of the people belong to category 2 and it takes a smart person to create something like this alone, my statement is not an insult.
Impressive. Soon after the mouse explore the maze, it immediately integrates the path to find shortest path based on its experience. As a neuroscientist studying spatial navigation, I'm deeply curious about what kind of learning algorithm they(he?) used.
@@suhu9379 @JUING-HUEI Thank you for answering my question! 😄👍 We know that there are specific cells that encodes specific location, direction, borderness, ect., but how brain integrate those informations to find a way from one place to another is still unclear. If I read your paper(I've found some) and I could understand it(hopefully), it may give some interesting improvement on my project. Again, thank you, have a nice day and happy new year!
@@brucewayne4036 What you are saying is roughly correct, however, human memory in general functions quite differently from computer data storage, even though both hold information. It is good to be careful about the computer-brain metaphor. They are very different things.
First run: around 1 min, announcer didn't announce exact time Second run: 6.869 seconds Third run: incomplete Fourth run: 6.469 seconds Fifth run: incomplete I believe each competitor only gets five runs, bad luck that this guy had two incomplete ones.
It seemed like he was cranking the speed up and down, and cleaning the wheels as much as possible for traction, trying to see how greedy he could get. The last run looked like a last ditch crank to 11.
I think the first run doesn't get time announced because it's the robot learning the maze. That said, how fast the robot can learn the maze would also me an interesting metric in such a competition
@@heroclix0rz maze competitions should let contestants make the software only, and they load their program into the same hardware which is shared by all.
I agree. I don't code, but sure would like to see how that side of things works. There has to be sensors involved too. I wonder if there are location sensors in the floor of the maze? Or do they use vision sensors only? That would be amazing!
@@timkuitems4431 likely 3-4 distance sensors (I would guess ultrasonic, but laser distancing is also a possibility) corresponding to the cardinal directions on the mouse. The really impressive part here, imo, is the remarkably precise control loop running the actual motors.
The little micromouse is just like me whenever I play a new metroidvania: starting slow exploring every possible paths and even going back couple of times where I've been, and later just skipping every area like if I live there
I don't know why yt recommended this for me. I dont even know this type of games are there. What an intelligent programming. Happy with this recommendations
@@joegerkrep7727 Did you seriously think that someone who suggested to use Dijkstra's algorithm wouldn't know that? Jesus FC, modify it to include parameters.
@@justincase4812 It's never nice to see someone being stupid, but it's definetly worse to find proud and stupid. Maximum speed gained from a particular run of edges is not a parameter of a particular edge. It is still not nessicarily a parameter of (strictly) previous edges (from a fixed starting point), so there's no way dynamic programming can save you in this case - you'd have to compute the maximum speed with a *new* search from each edge (rather than the origional starting point). If it wasn't bad enough already, you *still* couldn't even get away with pre-computing all the max-speeds at each edge (in a manner that would be useful), because each max-speed at an edge is dependent on your previous path, so not only are you doing a tree search at every edge, but you're doing it for every *path* to each edge, so there goes any DP potential you had. With some reasonable heuristics you could probably get that tree search at a particular edge to be constant time, most of the time, so you'd still end up in Θ(n^2) for the entire thing, albeit with a constant factor speed decrease over typical dijkstra that would be bad enough in 2D, but would triple(?) in 3D, as you're searching 26 directions as opposed to 8. Point being - the notion of using a greedy algorithm with a parameter that isn't trivial only ends up with the combined drawbacks of BFS and DFS with the benefits of neither. I'll give you credit for novelty, this is the first time i've heard someone suggest putting "parameters" in this fashion on Dijkstra, but lmao I hope this will also be the last. if you find yourself googling "best algorithm for (task)" to back up some stupid idea you didn't think through, don't be rude to people like joeger who clearly know more than you and just take the L.
@@justin9571 Excuse me? What a vitriolic display of pedantry - a complete hypocrisy on what you are hilariously accusing me of. But if that wasn't all.... No one is talking about dynamic programming, no one mentioned any of that which you have gone WAAAY out of your way to introduce strawaman after strawman. You must love talking to yourself. I don't have as much time as you do to write such a shameless vitriolic essay, however it is absolutely possible to determine max speeds at every traversed path. How can you possibly not undertstand that whether in 2 or 3-space. You ought to be ashamed of yourself.
I thought at first whatever it was doing during the first runs is just wasting time. But i think that it being slow in the beginning was on purpose while it was mapping out and saving infos about the corners and trying to calculate the best route to the yellow box. In the end, when it was re-released into the maze, it had the fastest way figured out.
@@suhu9379 thanks very interesting approach. One more question, how does it create its own reference points or local map? Does it purely rely on its on-board sensors with a dead reckoning approach? Or does it have a "global" reference like ultra wide band system or something similar?
@@pic18f452 Array variables are used to store the maze wall configuration. The starting cell for micromouse contests is always located at one of the corners of a maze, and the goal area will be at the right hand side when you stand near the start cell and facing the maze. This means that you can use two dimension coordinates to locate a maze cell, which is 18x18cm or 9x9cm. For classic micromouse mazes, which is 16x16 cells, we will set the coordinate of the starting cell to (0, 0). The goal area consists of four cells which are (8, 8), (8, 9), (9, 8), and (9, 9). For half size micromouse contests, the goal area will be designated before the contest begins, such that we can put them into the algorithm.
How much can cost a micromouse like this? Can I buy one in a kit?? I really want to learn how to do this and that fast :) Very good video. Thank you very much.
The entire project could cost under a hundred to several hundred dollars depending on what corners you are willing to cut, parts you use, experience and frugality, etc. The components list looks something like this: high traction tiny wheels (drivable axels included, presumably), motor/motors, at least 1 distance sensor array, a color sensor to determine when you've found the goal/start, some mechanism by why you will turn on a dime (preferably this means reversible drive along each side of the mouse which would be easiest to accomplish with 2 reversible motors), battery, wiring, microprocessor, emergency buttons for cutting power, resetting logic, etc. If you are a cheeky mofo, you'd say "ok, my first run will right turn + gap skip rules the whole maze to map it out, then my second run will calculate the exact trajectory needed to fling the mouse directly from the start to the goal and back to the start using a spring loaded chassis.
wait.. what is this game called..? i ve never see this before.. what is that mouse thing, a robot? how its go and came to the same place automatically..?🤔🤔🤔
When the mouse comes back to the starting position, it would either be confident in finding the shortest path or just finish dashing from the starting position to the goal area. That’s what we thought at the contest. You give me another viewpoint, which is also great!
Cave mapping is more difficult I think because of its arbitrary shape, infinite size, and 3D information. Micromouse map is limited in its size and confined in a 2D space.
The firmware programs are all flashed into a microcontroller on the robot, which includes motion control, path planning, maze solving algorithms. The mouse robot uses infrared sensors to find out what the maze configuration is and keep them in the memory. The sensors sense the environment from 1000 to 10000 times per second. Your may check the contestant's simple introduction about his mice in sites.google.com/site/ngbengkiat/Downhome/Topic1.
The participants would know before the contest date where the goal area is. For example, All Japan micromouse contest declared the goal area information on their webpage. www.ntf.or.jp/alljapan2022/
I find that alternating left and right turns at intersections will get you out of most mazes. I haven't found a maze that hasn't worked with yet. As far as finding a point in the center I don't know. My motivation if I'm in a maze, is escape.
We never tried this idea before 🤔. Whenever a dead end is met, we often use flooding algorithms to find a new path to the goal area, and continue to explore the maze.
Yes. The contest also serves as a target system so that students can try to integrate their knowledge and skills. They could also learn from competitive contestants those brilliant ideas in implementing the winning robot.
