Download "Averaging Gradius"

QuickTime Movie

Running Time: 2:57

Filesize: 26.4 Megs

Origins of the Project

Averaging gameplay, an idea which started from this discussion, sounds pretty straight forward - looking at how different players handle similar situations in a video game. What are the average actions that these players perform? What makes their run-through unique? What works best? How do people handle Bowser? What path do most racers take in Gran Turismo? What order do people take down bosses in Mega Man 2? How do people react to the The Sorrow in Metal Gear Solid 3?

The problem with these examples, and games like them, is that there's no easy way to compare the data. The player is usually free to move where and when they want. How do you find a way to analyze that, to draw any measurable parallels, when each player can move to completely different parts of the level at will? The answer is to choose a game which locks down those variables, to prevent that freedom of choice.

How about Gradius, for the NES?

Gradius, an auto-scrolling shoot'em up, is programmed so that the stage automatically scrolls at a fixed rate. With the stage constant, that allows us to compare how the player interacts with the AI and how, retroactively, the AI responds to that.

After I created a demo video as a proof-of-concept (which consisted of me playing a short section of the game several times) I invited people to send me gameplay recordings. I ended up with 15 different submissions, including my own.

The participants include all three main contributors of The New Gamer (G. Turner, D. Riley and myself), Nullsleep (you may know of his chiptunes at 8bitpeoples), four personal friends of mine, three members of the shmups.com boards and four from Insert Credit.

It is this selection of 'Level 1' runs that was used for the "averaging".

End Results

So what did it accomplish? What does the video show? Is it more than just a colorful kit-bashing of people playing Gradius?!

What first caught my eye when compiling the video, which came as a complete shock, was that the end boss wasn't set on a absolute path! Instead, it responds and reacts to the player's actions. Now perhaps this isn't a surprise to all those Gradius pros out there, but I was pretty impressed to see that sort of action from such an old game. In one of the tracks, the boss actually sits at the bottom of the screen, waiting to respond to the player's next move.

The average time taken to kill the end level boss was 20.055 seconds, with the fastest player finishing him off in a mere 10.01 seconds. Six people finished the boss off at nearly identical moments. It would seem that the boss, bored with the player, actually self-destructs after 27 seconds. Beyond the almost perfectly synchronized explosions, further proof of this self-destruction can be found in the videos: no 10,000 point bonus (given to players when the boss is defeated) was awarded to these six players and, in a few of the runs, the boss detonated when there wasn't a single bullet near it.

The infamous volcano sequence (which caused more problems and death with the crowd of submitters than even the end boss) certainly shows that the most common tactic is to park your ship in or around the top corner and blast away! Only two people did not go with this method: the red Vic Viper which sits underneath the lip of the volcano, and a blue one which actually floats in the middle of the action and yet never fires one shot.

Three people were not daring enough to go for the 5000 point bonus, and of the rest, most grabbed the bonus when the pathway through the mountain was the safest, when both exits were effective escape routes. There were two people, however, who grabbed it right away and one who waited for the very last moment to touch it (and allowed the screen to scroll him into the points.)

The average person prefers the Laser, at 2 to 1 odds; not so much the Double as only 2 people used it. Everyone used the Missile. Shields (marked as a '?' on the HUD) seemed to be actively avoided by some, as though it was a sign of bad form, and only a third of the crowd equipped it. Instead, most activated a second Option, that little glowy ball that follows the Vic Viper and fires, and it was this crowd who finished off the boss faster. There was a single person who had both the shields and two Options (at the cost of lasers.)

One person paused (thanks Mr. Riley, I needed the extra editing work).

But I found even the more pointless things incredibly interesting (and telling), like seeing when each person pressed the start button to skip the title screen from scrolling in, or watching as each Vic Viper, in sequence, would take out the red ships flying in a wave pattern, to leave behind power-ups in an almost perfect sine wave sequence. I love how the little mech-like gunpods together emerge from off screen, as a bright, white mass, and slowly break apart into a rainbow of mech clones.

Technical Garbage

Creating the video actually ended up to be a much more laborious task than I had expected. If you aren't actually interested in the process of creating this video, feel free to skip ahead to the next section...

So now that the cowards have left, let's continue. Originally I had intended on simply layering all of the submitted gameplay videos, tinting them a variety of colors to help the viewer differentiate one Vic Viper from the next, and be done with it. Visually the 'average gameplay' is represented pretty well, as the each of the Vic Vipers collide and sum up, they produce brighter and brighter whites.

It worked great in my demo. What I hadn't counted on was that the many layers of mountains and trees, since they are obviously present in each of the gameplay videos identically, would stack up and culminate as an over-saturated mass of white.

My first method of addressing the problem was simple. Too simple in fact. Instead of allowing all 15 tracks to add their hues up, only two of the fifteen were given this privilege, the rest merely laid on top without effecting saturation. This fixed the stage "white out" but introduced two new problems. One, the effects were, to put it bluntly, ugly. Secondly, it betrayed the goal of the project, namely averaging.

With two of the fifteen tracks being treated completely different than the rest of the project, it meant that those two tracks were inherently special. They interacted with the other gameplay videos in a different manner than the rest. It might have been a simple fix, but it was unacceptable.

My second solution, which sounds simple, ended up being amazingly complex but, fortunately, very successful. I needed to fix to original problem - the over-saturation of the mountains and trees. The obvious answer: create a 'plate' that would cover over the blown out stage graphics.

You know, if someone had just made a Game Genie code to remove all the enemies from the game, it would have been very easy! Or maybe if I knew how to hack a NES rom. But no, my fix had to been done in post-production! There are always more headaches in post.

Eventually, after some work, I got a video track that was just mountains and trees. I ended up combining a few different Gradius runs and keying out some colors to remove the Vic Viper, bullets, enemies, and all those other unwanted sprites. With my plate finished I could combine all the gameplay videos any way I wanted and just use the plate to cover over any ill effects that it might have on the stage graphics.

Now that I could complete the project the way I wanted to visually, it was time to tackle the sound.

A third setup was used to layer the audio tracks. I wanted to retain all the laser blasts, ship explosions, power-up tings and the other various sound effects, but I also had to keep the sounds from become too shrill. To that end, the gameplay audio tracks are each barely inaudible on their own, but when added up they create a fabulous sound. The final version sounds to me like it was recorded in a room where all 15 games were being played at once. The final version also includes an additional layer of music to heighten the score and some of the more vital (and special) sound effects are punched up a notch.

A fourth and final (phew) project was created to layer all the HUDs. The end result, while not very helpful in understanding who has what power-ups, at least hints at what power-ups are popular, with its ever changing hues of purple and blue.

The final product is a combination of those three secondary tracks plus the fifteen tinted Gradius runs, totaling about 43 gigs of uncompressed video.

Final Thoughts

The main goal was to present this idea of averaged gameplay in a way that people could digest. If the end product wasn't both entertaining and informative then what is the point?

Entertainment? Well, that's of course up to the individual, but personally, the final video surpassed all my expectations. I could watch the opening seconds of the stage repeatedly as the moment when all the various Vic Vipers splinter apart is just so mesmerizing.

In terms of information, it's there if you look for it. Statistically there's plenty to discover, my earlier examples are a small taste of what you can conclude from watching the video. But there's also a lot to find that just cannot be easily expressed in text. Watching how different players react to a spray of bullets; seeing how some go on the offensive and attack nearly all enemies while others fire less and dodge more; looking when certain people retreat to the back edge of the screen and when they charge forward; monitoring the enemies as they are destroyed, slowly peeling back the layers of color, possibly leaving a mostly transparent ghost to escape off the left side of the screen.

There's a lot to be said about how people play Gradius, at least a small group of us anyhow.

Miscellaneous Data

* Six people had this time. ** This player died partway through the level.