22.04.2022   Category: Site news

Episode 15: The ‘uncanny valley’ effect and VRC v4

Uncanny valley, what is that about, what does that have to do with VRC? Let me explain what it is why it is relevant for what we do, or what we have tried to achieve with VRC. A bit of background first.

This term ‘uncanny valley’ was first explained to me by my son Tim back in 2005, he was writing a thesis on human interaction with robots. For his experiment, he wasn’t using a realistic ‘human-like’ robot but rather an iconographic ‘cat-like’ robot called iCat who was interacting with people playing a simple card game. I believe it was a study for Philips Medical. Tim explained to me that using an iconographic robot was much more acceptable than trying to make the robot look and act like a real person and that this effect was called the ‘uncanny valley’.

This effect was first studied and published back in 1970 by Masahiro Mori, a Japanese robotics scientist. This was at the very early days of computer science and apparently, he was foreseeing robots already back then! He based his theory on earlier studies by Sigmund Freud about the psychology of the ‘uncanny’ or the ‘creepy’. He found a direct relationship between familiarity or acceptance and human likeness. This was gradual ascending relationship until around 75% human likeness, and then a sudden drop. This drop was reversed at around 80% and then a sharp rise to full ‘familiarity’ at 100% human likeness. This drop and rise in the graph is like a valley, that’s why he called it the ‘uncanny valley’, see the diagram. So, what does this have to do with sims in general and VRC in particular?

I touched on this subject in episode 5 when I was explaining about creating the right ‘feel’. In many ways a game or sim is subject to the same ‘uncanny valley’ effect. The ‘familiarity’ or ‘acceptance’ perception also has a similar relationship with ‘realism’ and in some cases ‘feel’. Very simple games like Mario Bros were so popular because they didn’t pretend to have any realism at all, they were well below the ‘uncanny valley’ drop-off. For many games it was mainly about the graphics and how well characters and sceneries could be animated. Graphics quality evolved with technology, animation quality evolved with software development and especially AI.

For a sim like VRC we are looking at 3 different areas where the ‘uncanny valley’ effect applies: graphics, sound, and handling. Each has its own ‘uncanny valley’ curve and has a different ‘weight’ in the total experience. Our aim has always been to create the most realistic r/c racing sim possible, with a realism of 95% or higher (although this is hard to measure of course)..

Graphics
About the ‘uncanny valley’ aspects of the VRC graphics, you either keep it simple or go all the way and avoid the ‘valley’! Our aim for v4 was to jump over the ‘valley’ by creating a state-of-the-art graphics engine that would produce jaw dropping graphics quality at a very high frame rate. Of course, easier said then done when starting from scratch! And with limited experience. I will write more about the development of the graphics engine in a later episode, it took us 6 years…

I knew already that Tony could deliver the highest possible track and car graphics, it was more a matter if our graphics engine would be able to turn Tony’s graphics artwork into a real-to-life experience on your computer screen, efficiently, and at a very high frame rate! With off-road on my wish-list for v4 we would be entering unknown territory. Thus far we had only done on-road tracks, developing off-road cars and tracks was going to be a whole new challenge and we had to make sure not to fall in the ‘uncanny valley’. And special effects needed for offroad, like animated suspension, dust, undulated tracks. For sure we would have to be at 90-95% on the realism-axis of the uncanny valley diagram for v4 to be a success. Could Todd and Tony deliver? Could Miro (graphics engine) deliver? A lot of uncertainties at the start of v4.

Handling
Handling or ‘feel’ is the result of a combination of the physics, the controller, and the graphics engine. How realistic will the car behave, how fast can you control it (respond to changes you see on the screen) and how fast can the graphics engine render the next frame in the highest possible graphics quality. The weakest link determines how you experience ‘handling’. The uncanny valley effect for handling is surely present. For a low-end game you can get away with low realism, in most cases you don’t even know how, whatever you are controlling, should feel. But if the graphics are such that you expect the handling to be of the same very high level it becomes critical, especially for users who are very experienced with r/c cars. They simply expect the car to handle exactly the same as their own car, even though they very well know that driving someone else’s car can feel totally different because of a very different set-up. What we have learned in the past years is that the upward ‘uncanny valley’ curve is extremely steep for experienced r/c racers. This probably explains why there aren’t that many top-level r/c racers in VRC, although we have seen top racers like Ronald Volker, David Ronnefalk, Martin Wollanka, Tom Rinderknecht, actively involved in VRC for quite a long time. Anyway, we fully understood that for v4 we had to up our efforts to reach a very high ‘handling’ realism. And now with off-road to be added, the bar was even higher.

Sound
The 3rd ‘uncanny valley’ factor is sound. You expect the car to sound what you are used to, whether it is a nitro powered car or electric. I already knew from v3 development that sound is a very complex issue, especially recording sound samples. The way Todd’s sound engine works, requires sound samples of 1 to 2 second length recorded at a steady rpm with no fluctuation. His sound engine then interpolates the sample with rpm. But under load the intake noise is quite different from an engine off-throttle. The same applies for electric motors, sounds very different on- and off- throttle!

And yes, we were also planning electric for v4, so even more challenges to conquer. At the outset of v4 I already had my doubts if we would be able to raise the sound-quality we had achieved in v3. At least I knew that what we had in v3 was acceptable, probably somewhere 85-90% on the realism axis of the ‘uncanny valley’ diagram, status critical I would say.

Here you have it, the challenges we faced with v4 were serious. Only time would tell if we could develop v4 to at least a 95% on the ‘realism’ axis of the ‘uncanny valley’. I was confident we could, but much would depend on the complete new graphics engine that Miro was going to develop.

That’s next.

Comments