I seem to recall the game "Bound", where you guided a low-poly ballerina through fairly abstract levels, had a really nice solution to "people don't bash into walls": if you tried to guide your character into a wall, she'd just start doing some stretches against it, which feels pretty accurate given how I tend to behave when I'm attending dance classes regularly. :)
Other characters might do different things; a mopey teen might slouch against the wall, a gun-haver might press against it and peer to the side, looking for cover, etc.
Obvious reasons not to do this start with "oh great, one more thing to add to the character state machine and debug" and "oh great, a few more animations to fit into the budget". Maybe also "it stops feeling like you're controlling the character". Someone whose day job is video games probably has at least a dozen more reasons, possibly even stories from when they tried it and abandoned it.
It's definitely possible to "not bash into walls"
You don't always need to have bespoke animations- sometimes keyframes and realistic constraints can get you there. I think Unity is working on a framework for natural human animation.
David Rosen pulled off quite a few little masterpieces -- his gamedev competition entries always had fantastically unique aspects. I cannot believe "Black Shades" was almost 20 years ago! (I have no affiliation, but I am glad to see https://www.wolfire.com/games is still alive and kicking!)
Lugaru and later Overgrowth have been the outcomes of what can only be called a decades long obsession with getting the player controls just right, allowing very natural and responsive control and movement, through creating emergent behaviors from first principles, resulting in a rich and natural interaction with the environment (and other characters, see the fighting mojo), in contrast to the many, many AAA titles that just constantly break the immersion through unnatural, jarring behaviors and frustrating limitations.
It's significant too: nothing frustrates players quicker then controls feeling "unresponsive" by doing something that's different to what they instructed.
Which is what some game does fundamentally bad. When using skills in some game. Your action got locked completely and won't respond to your request until animation finishes. (And when the mob hit you in the middle, you can't do anything because animation is playing.)
But some games makes more efforts into these and allow you to transition to another action in the middle of one, blend them seamlessly or even play them at same time. That makes the game feels much more responsiveness and enjoyable. Feels you are actually moving your body instead a online character with high action latency.
Is that really part of what uncanny valley represents now? I feel like we're moving the goal posts. Fifteen years ago the characters really could be uncanny and disturbing. I don't get that anymore. I feel like it's very close to real, but not real. However, it's in no way off-putting or weirdly creepy.
Watching the demo, I find it very impressive. Still in the valley, they key word being "still" - I think they managed to get past the bottom and are now climbing up.
That’s far from an unsolvable problem though, and this is a graphical tech demo, not a full game with the associated staff, budget, time, etc.
The Demon’s Souls remake (which I think actually looks nicer) has tons of destructible boxes, tables, stone work, etc (the original had much of this as well). I’m more interested high graphical fidelity being used in these smaller, authored spaces than in massive, often empty open worlds.
Filling up a small map with “real” content by hand is hard, doing it on a huge scale is impossible.
Still impressive technically and visually. Let's not forget that these things are running in realtime, they aren't a video. the whole "uncanny valley" thing is exaggerated. in the next 5 years you won't be able to tell the difference. In fact, you're brain will be constantly asking itself whether what you see is real or a 3D model. And I bet it did if you played the demo. That's a serious leap forward.
Now most videogames will still featured heavily stylized/cartoonish models, because some people are afraid of models being too realistic.
I get this in movies. Sometimes I even get this in scenes where they're using real actors and not CGI. Harsh lighting with make up can make humans look somewhat fake to me. One of the Mission Impossible movies gave me this effect.
When watching it I realized what you just wrote.
This is really impressive.
I spent a lot of time on both and read a lot about the issue, and most end up skipping physics on the PC, because it just feels worse controlling it realistically. This might be because we're used to that though. Anyway after all, you're making a game and games have to be fun to play.
This is what blew my mind the most with this demo, and I think will enable some crazy cool evolution in open world gaming.
Like in that video where you had to count how many times people passed the ball, but failed to notice a gorilla walked through the scene.
Let me first state that its amazing overall, very impressive.
Having said that, it exhibits pretty noticeable pop-in, also quite a few instances of what looks like garden variety z-fighting of textures/shaders in some places.
In a way these actually jump out at you a bit more in this demo than they do in less photorealistic rendered environments probably for reasons similar to the uncanny valley phenomenon being discussed throughout this thread.
E.g. the goal was probably to show off the new Nanite and Lumen rendering features, and probably MetaHuman for the main character faces and facial animations. At some point you have to say "ok, we focus on this stuff, and accept compromises for other things". If the team's goal would have been to create the best crowd simulation or environment interaction ever, they would probably have achieved this, but at the cost of other things.
Similar to ray-tracing, where you get great realism, but it is just an absolute resource hog. The number of vertices in soft-body physics animations come to mind. The real world has curves and deformations, and hair, with millions of strands, that each need tens or hundreds of physics-enabled vertices to look ultra realistic. Walking through a city, you might have 100 NPCs on the screen at once, who all have hair.
You simply need so much freaking compute power to pull this stuff off, I can't wait to see what it looks like in 10-20 years.
This is why I'm most excited for high-core counts leading to game development that allows these kinds of things to run in their own hardware threads. Imagine a world where each of those 100 NPCs has hair simulated on one thread, clothing simulated in another, world interaction on another, even AI generation on its own thread.
It's crazy just looking 10-20 years back and seeing where we are today. Compare this tech demo to Skyrim (2011) or GTA III (2001) and it's just incredible that we're looking at photorealistic, ray-traced, real-time "gameplay" like this on a home video game console. Or just compare GPU specs, the GeForce3 (2001), GTX 580 (2011), and RTX 3090 (2021) all have texture fill rates of 1.92, 49.41, and 556.0 GTexel/s respectively, with similar increases in raw floating-point performance. The future is exciting.
Oh, wait...
Also the acceleration of people and cars feels still too artificial. It is not the looks, but the physics that are still full on the uncanny valley.
I think as graphics get more and more realistic looks wise, game designers need to slow down the game, and make it feel more real.
One of my favorite car chases is from the movie Ronin. It feels extremely real, like you are there, as they didn't use any outlandish effects, but kept it grounded on how a real car chase would feel.
Super impressive demo though.
Who cares? It's a video game, that's how they work. I'd rather have developers work on making things fun than trying to make a creepy realistic Keanu not walk into walls
I suppose the tech can’t get across the valley without first passing through it
There's a lot of uncanny CGI in modern movies. There's also a lot of CGI that is so perfectly realistic you have no idea it isn't real. This is usually more environmental and less characters.
If the tech gets past the uncanny valley then you'll never know. This is basically true by definition. :)
No they’ll tell me
(And I’ll probably not agree, but if I do they’re still going to tell me, trust me they haven’t stopped telling me)
Aside from the motion though they look incredible.
You could animate that as hugging the wall, leaning against it or something, instead of a 0mph walk.
It is hard to stress how good and impressing that game was at the time.
At this stage I feel like the engine is very good and sure processing power is still a factor, but I feel like the biggest factor is cost/time. Small details that make something look natural are sure to go when there is budget or deadline.
That doesn't seem like much of a theoretical obstacle to me. The primitive controls only signal intent - you don't have to animate the human shuffling against the wall, you can just not move. The controller should be like whispering into the avatar's ear, not QWOP.
Or if you do want to grant the player lower-level physical control, you can have the human actually collide realistically with the wall. After all, people do bash into walls, just not commonly. You can even make them bump their head, and take damage.
"Even now people are confused, having watched the footage several times. “The first time you see the older Keanu that was a video shoot that we (Epic) filmed of him,” Pete explains. “After they come back after Morpheus that was all the CG versions. It was color corrected to match and look the same.” There is also one point where we see footage in the mirror this was real film footage that was comped in UE5. The first shot of Morpheus is also not CG, but then becomes CG when the young Thomas Anderson (Neo) appears. Even the desk scene at the beginning when Anderson is asleep was a real-time recreation of a scene from the original film."
Carrie Anne Moss, who on-camera looks stunning for her age....looks like she aged at least twenty years and her head was stretched, almost like they got the aspect ratio settings wrong. She's almost unrecognizable.
When they're in the car and the camera pans back to "IO" in the back seat, I felt like I was watching Who Framed Neo Rabbit.
Edit: here is a short video about the full size animatronic T. rex. https://youtu.be/fFTsYGgdR9k
Objects are often absolutely beautiful and realistic, but interactions between objects are ridiculous. Some of this is probably just "not enough processing power", but it is also due to where the rendering bumps up against the primitive controls. For example, you have a very realistic human. But then the human walks against an object, and she does this standard video game 0 mph duck-walk. Part of the problem here is that real humans don't bash into walls. So how do you realistically represent that, when in fact it's just Johnny pressing forward on his controller?