Motion capture video of pinball

[3waycombo]

I would like to suggest that you fine tune the physics of Pinball Arcade to closely resemble the gameplay characteristics of the pinball tables located at PAPA World Headquarters, If there was a standard for how a table should play, this would almost certainly be that standard. These are the machines that the top players in the world do battle to crown a world champion each year.

Plus they have gameplay videos for pretty much every game on the pinball.org website. They shoot these videos at a perfect angle to extract all kinds of data about how the ball is moving around the machine. And the person playing is no slap happy amateur. You're basically watching an expert at work, one of whom is Bowen Kerins, a former world champion, pinball ambassador and a really cool guy.

What's so important about watching someone like Bowen play is that you will see advanced flipper skills that show what a pinball flipper is capable of. To take it a step further, the ball data could be correlated with the way the player presses the flipper buttons and physically interacts with the machine. I think that incorporating this level of detail in the way the player interacts with the machine and the way machine interacts with the ball is critical in achieving a truly authentic video game experience.

All sports video games use motion capture video to recreate the way athletes move. I don't believe pinball is any different. I have no doubt that using motion capture video of a world class pinball player is the best, if not the only way to accurately reproduce the pinball experience on the computer.

 

[pinballchris]

For us mobile users, there is only so much you can do with the press of a screen without adding more places to touch. With the addition of the nudging options, I can't see adding more. I don't even use half the nudge options as it is. For PC and PS3/4, maybe.

 

[3waycombo]

I wouldn't necessarily focus on adding or changing controls, but using motion capture to accurately reproduce the way ball moves, most notably the way it moves in response to the flippers. There are a lot of intricate techniques that do require moving the machine in combination with pressing the flipper buttons but I wouldn't focus so much on that. I'd just like to see the physics be directly related to real world data, which motion capture video would provide. I suppose in order to get a full understanding of how flippers interact with the ball and translate that to the software, you'd not only need to analyze a pinball wizard's game, but also really bad players and everything in between. I bet the software engineers could make some good use of all that data. If they're going to use it to make the live catches and the skill tactics realistic, I guess they should make the flipper fumbles feel like the real deal to eh?

 

[sellenoff]

I think it's a great suggestion 3waycombo; I would love to see the physics taken to that level of accuracy. I am guessing, however, it would be too cost prohibitive to do as it would require a tremendous amount of time for each and every table. As it stands, the release schedule is very demanding and Farsight has to try and strike a balance, which I believe they do very well. As a pure pinball lover, I would desire nothing more than to see the worlds most accurate simulation of pinball, but I would imagine it would take tons of time of analysis, coding, tweaking, and repeating until perfect. I don't think that makes financial sense. Of course, this is just my opinion, I'm not representing Farsight's opinions in any way.

 

[sotie]

I think it's a great suggestion 3waycombo; I would love to see the physics taken to that level of accuracy. I am guessing, however, it would be too cost prohibitive to do as it would require a tremendous amount of time for each and every table. As it stands, the release schedule is very demanding and Farsight has to try and strike a balance, which I believe they do very well. As a pure pinball lover, I would desire nothing more than to see the worlds most accurate simulation of pinball, but I would imagine it would take tons of time of analysis, coding, tweaking, and repeating until perfect. I don't think that makes financial sense. Of course, this is just my opinion, I'm not representing Farsight's opinions in any way.

I was thinking the same thing.

 

[Sumez]

Physics is simulation, motion capture is animation. There's no way to "motion capture" a physics engine. It's not just numbers and curves, it's pretty hardcore programming.

A video demonstration could be used as a kind of unit test to confirm that the physics flow like they are supposed to, but you can't create them based on such.

 

[3waycombo]

Awww you just shot down my whole idea!! You make a good point. I guess the idea that was bouncing off targets in my brain was similar to what you mentioned about using the video as a study reference. While I think you are correct in saying that motion capture couldn't be used directly to create a table physics engine, you could use the data to calculate the motion of the ball very accurately and extrapolate it.

Since we could have known constants such as gravity, table slope, ball mass, friction coefficients etc and motion analysis would provide accurate velocity, spin and trajectory, we'd have enough information to build a model for forces applied to the ball from the flippers, bumpers, kickers and other playfield objects.

The reason why i think it would be so helpful to analyze a highly skilled player is because they use such a variety of subtle flipper techniques that you would get a full spectrum of possible situations to analyze versus watching someone who just hits the ball away every time the ball comes towards a flipper.

It would allow a sofware engineer to address complex situations, like if two balls are captured on a flipper, or the conditions necessary to live catch a ball... I could provide data on how to do a frustration tilt after I lose my ball. I'd do it for science.

 

[3waycombo]

I am guessing, however, it would be too cost prohibitive to do as it would require a tremendous amount of time for each and every table.

I'd be curious to know what it would cost. I'm not an expert either. I'm pretty sure that in many ways motion capture animation is actually cheaper than having an animator use traditional methods. Of course, as Sumez pointed out, mo-cap is an animation technique, whereas this is a simulation software.

From what I learned in the 15 minute "making Pinball Arcade" video, the dude with the superstar sideburns basically manipulates the "gravity" on different parts of the table to fine tune the gameplay. In no way am I going to criticize their techniques. I have every table from Season 1 and enjoy them all (except Circus Voltaire). For the most part it works great. However, it seems like a subjective approach, so the game designer's signature will always be embedded in the code to some degree.

I honestly wouldn't even spend the time to vent about it if they weren't so close to getting it right. I think it's because they've gotten it to this level that I'm itching to see the full potential. It's a bummer that money ultimately decide where this goes. "Hey it's only pinball!"

 

[Mayuh]

Coming from both cgi animation and coding, I really doubt mocap would be any help compared to the cost involved. Not even for analyzing. I did a few flipperskills videos using two gopro hero 3s at 120fps. I was completely stunned what the silverball actually was doing. It would be quite interesting for the programmers too, and very cost effective.
But as long as some features like ballspin, etc. are not in the simulation/physics engine it can not be taken care of.
It's a simulation after all, and adding more and more capabilities and parameters that influence the ball it will get even more realistic. FS is doing a great job on certain tables, very good job on others and on a few (older) tables it is ok.

I have faith in them, that they know what's missing - and with more powerful devices, they will add more features that will add to realism. Given that they are located in a winter sports region, there might be a few gopros around from their employees too

Regarding live catches... I'm able to do live catches on real machines and I tweaked some VP tables for days to be able to do them too. But the flipper's speed was too low then... From analyzing the slomo videos I still don't really know what's going on recoil, rubbers, axle bearing, coil strength etc. really hard to code that

Demo man: 'Close enough!'

 

[pezpunk]

I agree with some other posters: motion capture is not applicable to creating a realistic physics simulation. it's just playback data of some physics that happened. it's in no way predictive. you'd have to pre-capture every possible eventuality that could ever happen on a pinball table (good luck!).

the physics engine in TPA uses well-established and fleshed-out formulae and algorithms for predicting motion, but this kind of science is not an exact science. ALL calculations designed to model friction or rebound or aerodynamics or anything practical like that are inherently just models -- just an estimation, a rule of thumb. there is no fundamental simple formula like "area = pi r^2" for deducing friction, aerodynamics, or rebound on real physical objects in gravity and atmosphere. i mean you can certainly find formulae to describe these things, but they are just estimations of variable precision. the area of a circle is a theoretical thing, a fundamental concept. whereas friction, for example, is just our description of what we see at a macro level, which is really the result of dozens of different hugely complicated things going on, such as gravity, atmospheric pressure, imperfections in the surface of the ball and the table, magnetic fields, the vector of the ball relative to the center of the earth's mass, and so on.

ideally, Farsight could build a complete and working model of the universe and everything in it, in which to place its pinball machines. that would be the most accurate way to properly model ball movement. this approach would require their developers to solve a certain number of outstanding cosmological issues regarding newton's view of the universe, figure out how gravity works exactly, and would likely require significant advances in our understanding of quantum mechanics as well. stephen hawking could be brought in on a consulting basis.

but lacking that, what game companies generally end up doing is developing algorithms that approximate the practical upshot of how a ball tends to move in the real world. these algorithms aren't laws, they are just models that resemble the working world at a macro level.

and oh yeah -- also the calculations have to be rudimentary enough that they can be performed by an iPad in real time with whatever cycles are left over after it's rendered all those polygons.

 

[brakel]

The number of calculations required is also a limiting factor. I forget if Mike said that the physics engine is calculating the ball location 60 times per second or 120 but it needs to be able to keep up with the FPS of the graphics at a minimum or weird things will happen. Mike said that the consoles are able to do that with their current physics engine but that the mobile platforms sometimes had issues keeping up. Reading between the lines it sounds to me like the current consoles can't be pushed much more. There is hope though that the next gen consoles and the PC releases could get some physics upgrades. They've hinted at that for the PS4 release. The other good thing is that mobiles are getting faster CPUs with every new hardware release. Eventually I see them being able to upgrade the physics for the mobiles as well.

 

[Mike Reitmeyer]

The number of calculations required is also a limiting factor. I forget if Mike said that the physics engine is calculating the ball location 60 times per second or 120 but it needs to be able to keep up with the FPS of the graphics at a minimum or weird things will happen.

Just for clarification. Physics are updated 60 times per second on all platforms as a minimum. In some cases (ball near the flipper) we update 360 times per second for more accuracy. Speed of the ball also varies the update timing. If the ball is moving really slow, 60 times per sec is good enough, if it's moving really fast, we may doe 120, 180, etc.

And as pezpunk said, we don't use accurate real world equations cause those would be too costly and too hard to create and tune. We instead write up equations that give us as accurate model as we can get without killing the performance.

There is a similar issue with collision models. Those can only be partially accurate to the table (curves are the issue mostly), because the more polygons we used to represent a collision the harder it is on our collision detection code. So again we try to find a good balance of speed and correct look/feel.

The final limitation we have is input is only collected 60 times per second (possibly 30 in old phone hardware, not sure). This is a hardware limitation we can't get around. This makes special flipper moves harder to simulate because the timing of the input is a little off.

That all said, we are always looking to ways to improve the physics.

 

[Sumez]

Context sensitive increase in physics and collision calculations, I'm glad to hear you are doing this!

As for curves, have you tried actually making code that can calculate the collision and physics in relation to a single curved edge, rather than creating "fake curves" out of a lot of tiny polygons (as in the actual graphics)?

 

[qwertydvorak]

Not a full mocap idea but maybe a pinball with an rfid chip inside. Put readers in each of the outlanes and the center drain. Let real people play the machines from average up to pro. Complie the data and compare to a roughly equal number of gameplays on the same table in TPA and if the numbers don't roughly match tweak the physics or the table to correct. Do the same type of thing with tables that have ramps or other certain areas prone to STDM drops. A perfect example of the second one would be Black Knight and the center ramp that drains STDM on the start of 2 ball multiball EVERY SINGLE TIME.

 

[3waycombo]

Ok so the consensus is that modeling a simulation based on actual physical universal laws is impractical (at this time) mainly for three reasons. It would cost a lot. It would require extremely complex code and it would require too much computing power, especially for mobile devices. At least nobody said it's impossible!...

At the very least I think this makes for good discussion. Thank you all for your comments. In particular to Mike at FarSight, I appreciate hearing what you have to say. I think the Pinball Arcade is awesome and a great value. It's not easy to produce a high quality product at a low price. I'm excited to see what you guys have in store for us in the years to come....

Now I have to go research some of this stuff because you guys dismantled my idea pretty thoroughly.

 

[3waycombo]

Not a full mocap idea but maybe a pinball with an rfid chip inside.

This is actually along the lines of the way I'm thinking about this. I don't know what sensors would be required or any of the technical stuff. I'm picturing a pinball machine set up with a system of sensors which a computer analyzes game after game. The analyzing computer would likely need to have the ability to learn so it could become better and better at predicting the motion of the ball over a wide variety of situations. This technology already exists to make a robot that can walk on two legs. Apparently it's almost impossible to build a robot that can walk right after it is built. Essentially it must be programmed to learn how to walk by practicing.

In much the same way, a pinball-analyzing computer could become better at predicting the way a ball will behave as it accumulates data over time. This computer wouldn't necessarily have any understanding of physics, but ideally it would be an automated system that output it's "physics" model in the form of usable code.

I don't think that this would require any technology that doesn't already exist. Of course it would require a large investment and a lot of really smart people to create in the first place. However, once the system is operational, it would pretty much run itself. All you'd need is people to play pinball constantly. I think that would be a great job for me, so let me know when you're ready. I'll be there.

 

[brakel]

This is actually along the lines of the way I'm thinking about this. I don't know what sensors would be required or any of the technical stuff. I'm picturing a pinball machine set up with a system of sensors which a computer analyzes game after game. The analyzing computer would likely need to have the ability to learn so it could become better and better at predicting the motion of the ball over a wide variety of situations. This technology already exists to make a robot that can walk on two legs. Apparently it's almost impossible to build a robot that can walk right after it is built. Essentially it must be programmed to learn how to walk by practicing.

In much the same way, a pinball-analyzing computer could become better at predicting the way a ball will behave as it accumulates data over time. This computer wouldn't necessarily have any understanding of physics, but ideally it would be an automated system that output it's "physics" model in the form of usable code.

I don't think that this would require any technology that doesn't already exist. Of course it would require a large investment and a lot of really smart people to create in the first place. However, once the system is operational, it would pretty much run itself. All you'd need is people to play pinball constantly. I think that would be a great job for me, so let me know when you're ready. I'll be there.

You can capture all the data from every pinball game that's ever been played and it isn't going to help write the code for how the ball should react when it hits something. I think anyone that has played real life pinball for a few years can look at a simulation and tell you whether the look and feel is right or not. I don't need a computer to tell me that.

 

[3waycombo]

You can capture all the data from every pinball game that's ever been played and it isn't going to help write the code for how the ball should react when it hits something. I think anyone that has played real life pinball for a few years can look at a simulation and tell you whether the look and feel is right or not. I don't need a computer to tell me that.

Just about anyone can tell the difference between a pinball video game and the real world, but I doubt many people could tell you with much accuracy how a pinball moves around the table without taking measurements. Observation is not quite the same thing as measurement.

 

[superballs]

I think it's a great suggestion 3waycombo; I would love to see the physics taken to that level of accuracy. I am guessing, however, it would be too cost prohibitive to do as it would require a tremendous amount of time for each and every table. As it stands, the release schedule is very demanding and Farsight has to try and strike a balance, which I believe they do very well. As a pure pinball lover, I would desire nothing more than to see the worlds most accurate simulation of pinball, but I would imagine it would take tons of time of analysis, coding, tweaking, and repeating until perfect. I don't think that makes financial sense. Of course, this is just my opinion, I'm not representing Farsight's opinions in any way.

I don't think that video should be used for physics of course but, as much as I love TPA, there's something off with the way the ball comes off the flippers. It also doesn't lose upward and gain downward momentum enough, so the forces of positive and negative acceleration (non lateral). A ball travelling laterally will accelerate downward much quicker IRL than in TPA. Also, the ball comes off the flippers with a lot more SNAP.

This is something that seems to affect all pinball simulations. I've yet to really see this modelled exactly, I'm just using TPA as the example as it's my primary pin sim. This also isn't a game destroying flaw as most of the tables seem to play very well despite this.

The final limitation we have is input is only collected 60 times per second (possibly 30 in old phone hardware, not sure). This is a hardware limitation we can get around. This makes special flipper moves harder to simulate because the timing of the input is a little off.

That all said, we are always looking to ways to improve the physics.

I'm going to assume that you meant CAN'T GET AROUND there, careful there be rabies round these parts.

 

[Mike Reitmeyer]

I'm going to assume that you meant CAN'T GET AROUND there, careful there be rabies round these parts.

Fixed that, thanks for pointing that out.