Rodrigo

So I've been working on a big gamepad patch for a bit, and now that I'm getting further into it I want to work in combo/compound controls. Similar to the fireball/dragonpunch type combos from Street Fighter.

It took me a bit how to work this out as I hadn't dealt with specific sequence of events in Max before, and I got it working in the end, though I'm certain there has to be a better way. There's also the additional problem that my fireball combo gets triggered when I do a dragonpunch combo too. I can add an additional state/step requiring both buttons to be 'off' before a combo starts, but I think it might just be better to refactor it altogether.

Here is my code for a fireball and dragon punch:

And here's an example I found online on how it's actually programmed:

The following example shows how a fireball combo move is triggered by pressing the navigation keys for "down, down+right, right", in this order.

ComboMove fireball = new ComboMove("Fireball");

fireball.press("Down").notPress("Right").done();

fireball.press("Right").notPress("Down").done();

fireball.notPress("Right", "Down").done();

And here's an example I found online on how it's actually programmed:
The following example shows how a fireball combo move is triggered by pressing the navigation keys for "down, down+right, right", in this order.

ComboMove fireball = new ComboMove("Fireball");
fireball.press("Down").notPress("Right").done();
fireball.press("Right", "Down").done();
fireball.press("Right").notPress("Down").done();
fireball.notPress("Right", "Down").done();
fireball.setUseFinalState(false);


how-to-program-sequenced-events-ala-video-game-combo-moves

Here are both combos made such that they can't trigger each other (by checking for both buttons being off first).

I don't have a definitive answer, but you can encode the state of buttons as binary digits and then check for proper sequence with [match] object.

That's much better using match. I experimented with coll to reduce the lists down to individual characters but didn't know what to do with them after that.

Will make it easier to tweak/add/modify the combos too (ie skipping the down+right bit and only having down and then right for doing it really really fast).

Here's a variation using bucket that might be worth looking at.

Ok, here's a solution using coll so I can just type in combos using the actual words (ie "nothing down down+right right").

Ok, here's a solution using coll so I can just type in combos using the actual words (ie "nothing down down+right right").
(use arrow keys on keyboard to control)

It just occurred to me that I might want to expand this to allow for sequences to happen within a specific time period (allowing for wrong presses). Basically an amount of 'slop'.

Say I want to do a combo of "left nothing right nothing left nothing right", it's quite possible that I might accidentally hit an "up" or "down" in the mix somewhere. A similar problem might arise with any longer sequence, that it may not be EXACTLY right. I could implement wildcards into [match] but I wouldn't know what order the 'wrong' messages may come in.

So looking into it further it looks like its a finite state machine that's generally used, which is pretty much what I've built.

Where I run in to trouble is how to deal with accidental input in the middle of the state machine without resetting back to the initial state. To basically allow for a sloppy finite state machine where all the states would still be needed to happen in order, but with the possibility of interruptions.

I also thought about using a neural network/machine learning algorithm, but I want to try to solve this problem without such heavy lifting.

Found this page which gets into more detail but it's still quite 'exact' state machine built.

http://gameprogrammingpatterns.com/state.html

So looking into it further it looks like its a finite state machine that's generally used, which is pretty much what I've built. 

Found this page which gets into more detail but it's still quite 'exact' state machine built.
http://gameprogrammingpatterns.com/state.html


maybe i misunderstand, but that kind of "sloppiness" sounds maddening? what if you changed your mind and didn't want to do the combo and do something else but it predicted you were going to do it and executed the combo? sounds like an interesting interaction experiment, but i imagine it might be kinda like someone that keeps interrupting to complete your train of thought.. wrongly.

maybe allowing for slowness in action or allow for repeated measurements in the same direction? like if they were holding a button during multiple sampling times. for instance, if down, left, up was the combo, down....left....up or down, down, left, left, left, left, up would work. still, i'm not sure that would improve the experience.

games that i'd rate as sloppy in the good sense generally seem like they actually have super high sampling rates and are very sensitive, so that if you think about your actions you're kinda wobbling around like jello and tend to overshoot or undershoot your movement, but if the game makes it compelling enough for you to put in the effort to do without thinking they feel both super tight and fluid. lol

if you want to accommodate people making the wrong combos, i'd just make the combos easier? :)

imo the most important thing is getting low delay feedback of successful action. this is an interesting test: 

http://www.csfieldguide.org.nz/_static/widgets/HCI/HCI-Delay/index.html

 for me below 50ms seems more or less instant. i suspect you could get away with more in higher action situations with multiple simultaneous events going on, where you couldn't pay attention as closely but i'm not sure if that wouldn't just make the interaction feel "wrong" without being able to put your finger on it, so to speak. :) audio is different though...

http://www.gamasutra.com/view/feature/132122/measuring_responsiveness_in_video_.php

http://www.eurogamer.net/articles/digitalfoundry-lag-factor-article

http://enterthesingularity.blogspot.com/2010/04/latency-human-response-time-in-current.html

(spoilers: to measure the accuracy of the input they train a camera on the controller, to make it more accurate + leds. i've never tried to do gesture recognition in jitter... might be interesting to try and train your button combos against gestures? or use jitter to troubleshoot that the buttons coming in match what you are doing on the controller?)

imo the most important thing is getting low delay feedback of successful action. this is an interesting test: http://www.csfieldguide.org.nz/_static/widgets/HCI/HCI-Delay/index.html for me below 50ms seems more or less instant. i suspect you could get away with more in higher action situations with multiple simultaneous events going on, where you couldn't pay attention as closely but i'm not sure if that wouldn't just make the interaction feel "wrong" without being able to put your finger on it, so to speak. :) audio is different though...

these articles might be of interest too.
http://www.gamasutra.com/view/feature/132122/measuring_responsiveness_in_video_.php
http://www.eurogamer.net/articles/digitalfoundry-lag-factor-article
http://enterthesingularity.blogspot.com/2010/04/latency-human-response-time-in-current.html
(spoilers: to measure the accuracy of the input they train a camera on the controller, to make it more accurate + leds. i've never tried to do gesture recognition in jitter... might be interesting to try and train your button combos against gestures? or use jitter to troubleshoot that the buttons coming in match what you are doing on the controller?)


From the beginning of input you would have a growing list of entries.

These could be compared to a list of matches. In this process one challenge is to "scan" through that list (because there may have been a wrong entry), basically setting the start-point for the comparison further and further up the sequence.

A test of "I want x percent of list a to match list b" would be possible too, to deal with wild cards, would it not?

Fuck, I had typed out a whole long response and then the c74 forum took a dump. Balls.

Ok, so in terms of clarifying 'slop' it's more about the tolerance of stray events. So in trying to do a fireball combo (down, down+right, right), I may accidentally just do down, right, or back, down, down+right, right, etc....

In playing street fighter (and other games) back in the day, I felt that I could very consistently trigger the combos, but in programming them in Max, it 'feels' less accurate. Like harder to consistently trigger, so I'm thinking there's a different programming paradigm underlying it.

Yes! Latency/timing is crucial in this kind of context and something I'm very aware of. So far I'm leaving primary functionality for regular button/trigger/analog moves, and using combos to engage secondary processes/functions that are less timing critical. That may change for certain things, but working on one thing at a time.

I think I might try doing some of the machine learning stuff for this to see if that fare better. Had a good response in the machine learning thread from Ali with regards to this type of usage (Wetterberg might it interesting too as a fellow machine learning noob!). Looks promising.

https://cycling74.com/forums/machine-learning-in-max-hallelujah/

I also tried the ftm gesture follower external, and couldn't get good results out of it since I couldn't figure out a way to give it more than one input sample to test against, which isn't ideal for the dpad type boolean data stream.

Ok, so in terms of clarifying 'slop' it's more about the tolerance of stray events. So in trying to do a fireball combo (down, down+right, right), I may accidentally just do down, right, or back, down, down+right, right, etc....
In playing street fighter (and other games) back in the day, I felt that I could very consistently trigger the combos, but in programming them in Max, it 'feels' less accurate. Like harder to consistently trigger, so I'm thinking there's a different programming paradigm underlying it.

@Jonah
Yes! Latency/timing is crucial in this kind of context and something I'm very aware of. So far I'm leaving primary functionality for regular button/trigger/analog moves, and using combos to engage secondary processes/functions that are less timing critical. That may change for certain things, but working on one thing at a time.

I also tried the ftm gesture follower external, and couldn't get good results out of it since I couldn't figure out a way to give it more than one input sample to test against, which isn't ideal for the dpad type boolean data stream.


How to program sequenced events (ala video game combo moves!)