> The animations look so fake.

Let’s talk about why it looks fake.

I think it comes from the use of hydraulics. Biological muscle is way less linear (it works more like a second-order control system, where things often overshoot their target before dampening).

But one part of it may also be algorithmic: the robot’s Kalman filter (?) may correct for errors from sensor fusion much faster than a human would.

I would love a more informed opinion than mine though.

This video has very little motion blur compared to what you normally see in high quality 4K 24/30 FPS footage (e.g. Hollywood movies). That contributes to the uncanny look, giving it an "animated" or "sped up" feel. Which is funny because most robot demo videos are sped up, but not this one.

It can't just be hydraulics though, because to me the most telling moments that look fake are when the robots are jumping - they have too much hang time/too much uniform velocity for the relatively small amount of vertical distance they get. In particular, see the initial jump at 15 secs in the video. It's like the robot is floating instead of jumping.

I had that same distinct impression about the jump at :15 the first time I watched, but watching it again i don’t. I think it has something to do with the leg motion while airborne. If you just focus on the torso it seems normal.

They must be heavy, which helps with fluid motion

To me it looks like claymation. But of course I can't really explain what it means for something to "look like claymation" to me. Is it that the movement is too isolated? Not isolated enough? Both of those things at the same time in different ways?

I agree but I chalked that up to motion of extremely rigid bodies under large torque.

Watch the reflections on the plexiglass while they are moving. Obviously they could have been composited in but there's definitely a live portion of the shot.

Subliminally jerky - you can’t directly perceive it but you are aware of it.

"Subliminally jerky" - When you're certain it's dehydrated meat, but you don't know why.

It looks like 'fake' and animation because it _is_ an animation. The dancing was likely motion captured and/or hand animated and superimposed onto the robots basic balancing control system.

> onto the robots basic balancing control system.

I don't think "basic" is the right word.

I'm not sure what you are saying here. I think the intent is to differentiate between the "optimal" robotic control dynamics which is designed for safety, efficiency, and to respond to unexpected situations and "dancing" which is intended to either mimic human motion or some anthropomorphic version of it.

So it is "animation" of a physical sort designed to make it look like the robots are responding to the music naturally, rather than programmatically.

I think it looks fake because the movements are so smooth and fluid.

On a dancing person you see all the slight adjustments to keep balance that make it feel natural, I believe these robots do it too fast for us to notice and so it seems like CGI where keeping balance isn't a thing.

I'm looking at the jump at 2:09, and all I'm seeing is a super wobbly robot trying very hard to keep its balance?

Look at the pelvis movements at 1:15 to 1:25! You can clearly see it jerking around so the robot keeps its balance.

None of those movements are smooth or fluid, it's all jerky jittery and unnatural, which is exactly what you'd expect from actual robots. If it was motion-capture, you wouldn't get those movements, you'd get smooth human movements instead?

I'm so fascinated by how this video triggers people's "it's fake!" senses, when it's clearly real.

Hydraulics make robot motion look fake? Have you watched real robots move before? I've never once watched a robot move and thought to myself "that looks fake".

This is how the robot moves IN REALITY: https://www.youtube.com/watch?v=uhND7Mvp3f4

Oh, man. Hard to watch the sequence starting at 1:56 without wondering when the robot is gonna take that guy's hockey stick and beat him to death with it. Do you want a robot uprising? Because this is how you get a robot uprising.

It can look real with hydraulics, but it can look fake.

In that video you linked, look at second 21/22, when it raises the arms in the air: the way the arms stop instantly is very inhuman.

For me, the front toe taps are especially fake-looking. It’s hard to pin down exactly without going frame by frame, but to me it feels like it‘s a combination of the lighting and the way the foot “rests” on the floor (which should probably be deforming since it looks like dense foam?)

Almost as though they had to edit out some balance-assisting device.

Also: why are the people at the right speed, but it feels like the robots are sped up? Them being sped up would be fine, but if the two are at different speeds then it’s a clear indicator that it’s composited.

Agree that if anyone except BD put out this video I’d bin it as fake immediately. I still have a hard time saying I believe it - it feels like such a jump from where my mental model of their robots was and CGI is so good at this point that it feels easier for this to be a fake than real.

It might help to compare against video with less editing and effects. Several youtubers are playing with Spots and you can get a better idea of what they're really like.

Adam Savage: https://youtu.be/-R8wUybrspo?t=2374

Marques Brownlee: https://youtu.be/s6_azdBnAlU?t=249

They're only like what they're "really like" because in those videos they're being operated from behind a layer of abstraction that only allows a few really basic movement patterns (go X direction, at Y speed, within allowed parameters).

I think this new video is trying to illustrate more of the highly dynamic capabilities of the machines that can be used by more specialized operators and software. (And I don't think there are any "effects" exaggerating those capabilities like you're implying, unless you mean things like color correction or the occasional cut to another piece of footage).

(A Spot, for those—like me—who didn't know, is the yellow 4-legged robot.)

It does, but counterpoint: unassisted (no help balancing) bipedal motion has been a holy grail for a looong time, where 4-legged is child's play by comparison. This could mean that just the bipedals are faked.

It would help a lot more to compare footage with real-world people interacting with the bipedal robot.

>It would help a lot more to compare footage with real-world people interacting with the bipedal robot.

Maybe I misunderstood your comment but there are definitely videos of people interacting with Atlas (the bipedal robot) to the point that it's a running joke that they'll remember said interactions. See https://youtu.be/rVlhMGQgDkY?t=83

Same, and I've been working with some of the latest and greatest collaborative and industrial six-axis and SCARA robots. There's definitely a back and forth between innovation and legacy in this industry. For example, FANUC famously refuses to deprecate or drop service for any of its equipment, you'll pay through the nose for some of it but they're willing to keep you supplied with parts, service, and bug fixes for 40-year-old equipment if you've got it. Also, a lot of it is clearly designed for "what's the most we can get an average maintenance tech to understand after a 40-hour certification class", not for programmers and engineers. But with a demo like this, it's obvious that BD is working on a level that the industrial automation competition won't reach for decades.

Please tell me you have a blog

I don't, but I do have a domain that's sitting idle except for a mail server. Maybe a blog could be a theme for 2021...

I had the same feeling, and I assume they evoked this feeling intentionally trough the lighting, camera movement, etc...

I felt the same way, but I think it might be because of the context that we're used to seeing advanced robots. Currently, that context is video games and movies, where we know they're fake. In a sense, we're conditioned to see this kind of tech and expect that it is done with CGI, so it makes unbelievable when we see it done for real.

That's why Boston Dynamics should put on a live show.

Cirque du Robots

sound would help, and maybe someone can dig up the real audio version if it is available. because hearing those engines roar and servos buzz, you appreciate there is indeed more going on than meets the eye

> aggressively smoothing the motion since mechanical systems don't have the ability to accelerate & jerk as fast as biological systems

I wondered if they covered the bots in a flowing dress if it would sell the illusion even more.

You might be interested in a couple of videos by an indie vfx group called Corridor

https://www.youtube.com/watch?v=dKjCWfuvYxQ https://www.youtube.com/watch?v=y3RIHnK0_NE

Mandatory XKCD reference: https://xkcd.com/331/

I would go further than that and say that it's obviously CGI, though I don't know if that actually means it is CGI - it could be, as you mentioned, some sort of post-processing. Or maybe it's greenscreened and composited together?

More than the smoothness, there just seems to be something off about the weight and momentum that you tend to see in high-end, big-budget-movie CGI, but it's hard to put a definite finger on it.

There's a similar effect in old movies when a super-strong character picks up a massive boulder that is actually made of styrofoam or something. You can tell that the mass is wrong based on the way it moves.

I think this is a similar effect but in reverse -- we know the limbs are heavy and aren't made of styrofoam, so the fact that they are moving like they are is uncanny. Our brains aren't used to motorized and hydraulic limbs.

You can get a little bit of this effect watching assembly line arms sometimes, but it's less jarring because they are clearly disembodied, whereas this thing is fairly humanoid.

Or even in modern movies with CGI-enhanced characters.

Gollum moves as if his enormous head was empty, because the actor doing the motion capture didn't wear a heavy helmet.

In "Rust and Bones", Marion Cotillard plays a character that loses her legs after an accident, and she has her legs removed by CGI, but there are scenes where she's carried by Matthias Schoenaerts that look unnatural, out of balance.

I see this frequently even in newer movies/tv--when actors are drinking out of empty coffee cups.

That's a great example. I wonder how much of that is due to the (lack of) the weight of the coffee, vs. not having to compensate for sloshing liquid that could spill if you aren't careful.

Could it be slightly sped up? The robots do make jumps so one could calculate it from that.

I was curious so I just timed the jump at the beginning of the song about 15 seconds in.

Assuming the jump is about 30cm, and I stopwatched it at 0.3 seconds from the apex to the ground, you arrive at an acceleration of 9.6m/s^2 which is basically spot on the acceleration due to gravity.

At 0:57 the camera pans to a position where you can see people on the upper floor walking by in the distance. You can see them move as the robot dance, which shows the video is not sped up. You could claim they were composited in (a suicidal move for BD, if they ever tried that), but if you look closely at the windows below, you can see the the camera crew in the reflection, who also move at the correct pace - and that shot would've been hell of a work to fake. In conclusion: this is real-time.

I think it's clearly recorded in-camera, and I think real camera footage - the reflections in the (plexi?)glass screens show the robots and people holding cameras. I'm not even seeing any obvious indications it's been denoised.

As someone who works in the VFX industry, faking reflections on wobbling/flexing glass screens (static ones are easy) is actually pretty difficult to do right, and I doubt they would have added the camera operators if they were faking it.

Some of the aspects making people think this is CG are the diffuse lighting (no obvious hard shadows or occlusion), very clean robots, "weird" (i.e. not normal, but I believe it's honest for the robots) motion, and lack of sound.

I think what happened is they used a fancy camera with a big lens and a 30 FPS frame rate, giving a very high quality image reminiscent of Hollywood films, but they set a faster shutter speed than Hollywood generally uses, giving less motion blur than is normal in high quality footage like this. The relative lack of motion blur despite the low frame rate gives it a jerky appearance similar to sped up or timelapse video, or animation. Combined with the slightly odd motion of the robots, it has an uncanny look.

It's kinda funny because most robot demo videos are sped up and/or CGI, but this one isn't.

You can see the vibrations of the plexiglass walls and the camera crew reflected in them vibrating in amazing detail. You can also see people walking around in the background in normal speed. It doesn't look like it's accelerated, or I would say no more than 10%. Also given the visible but subtle marks the robots leave on the mat, this is either next level CGI and compositing or next level robotics, and BD are not a next level CGI company.

I think it's absolutely fascinating that so many people, here and especially in the YouTube comments, are thinking this is CGI, when doing this with CGI would be much harder than doing it for real.

Also, if you've followed Boston Dynamics for a while, what this video is showcasing isn't that much of a leap forward, they're just iterating away on their robots, adding capabilities and movement patterns.

I'm inclined to agree with this after bumping the video up to max quality. I do believe that this is real, and you can see from 0:51-1:05 that there's some reflections going on that would be a lot trickier to fake. But this footage has a similar quality to what I see in a lot of heavily processed drone footage, where it just doesn't look like the same way I see the world.

Also watch their feet very closely at max resolution.

There's a lot of shock rattling and wobbling I've never seen in any CGI robot animation, not to mention they are leaving visible indentations in the foam floor. If it was CG they really went all out.

If you look carefully, you can also see reflections on the robot's visors.

Change playback speed to 50% and it looks more realistic. That's probably realtime and they sped up the video.

I don't know... watch the people walking in around 1:00 in the top left. Looks pretty much real time.

https://youtu.be/fn3KWM1kuAw?t=58

A simple test shows this isn't more realistic. Gravity during the jumps are off, and people walking upstairs is odd. It looks like it was filmed in realtime.

75% looks most similar to the previous Atlas videos. Also the movement of the people upstairs behind the glass or the reflections of the cameraman confirm that it has been sped up.

I don't know how fast you walk, but at 75% those people are definitely in slow-motion. I could see maybe 90%, but 100% looks completely natural to me as far as their movements are concerned. (Personally, I want to see a jump at 90% to judge, not someone walking.)