I can't believe no one's talking about Infinideck

"mptp" wrote:
It's technically either impossible, or ludicrously difficult. [...] Keep in mind we're skeptical with good reason, not skeptical because we don't like new technology. [...] the company is calling it the future of VR, without having fully explored its tractability

You probably didn't know, but implementations of this concept have existed and have been validated for the past 18 years and that's what people in VR call an omnidirectional treadmill. It's been created first for the US Army with a $400K contract in 1996.

The Virtuix Omni and the Cyberith Virtualizer took the name but they are only passive sliding surfaces, not omnidirectional treadmills.

The seminal paper that defined the term :
The Omni-Directional Treadmill: A Locomotion Device for Virtual Worlds (1997)

Example :



Third-generation (2009) : http://virtual-dio.com/previous.html#odt

Other implementations :
- Torus Treadmill : http://intron.kz.tsukuba.ac.jp/torustreadmill/torustreadmill_e.html
- CyberWalk : http://www.youtube.com/watch?v=d-8eVcN2z3k

I'm aware of their existence (and that the Omni and Virtualizer's aren't treadmills at all), but I've only ever seen solutions that are controlled by an external spotter.
The solution that the army purchased, from what I understand, was adapted to be a system in which the operator simply steps into the forward-area of the controller to move forward, etc.
Maybe they have a working system that is reactive to the user's movement, but I can't find any information, pictures or videos about it, except for that one photo which is shown in your link 'Third-generation'.

I'd dearly love to be proven wrong, but it seems like the idea's been around for a while, and nobody quite knows how to solve the core issue of 'how do we control the direction and speed of the treadmill?'.

Here are three use cases that would cause anything but the most perfect sensing/engineering combination to sweep the user off their feet:
1. The user is sprinting forward at full speed, then slams their front foot downwards to come to an immediate stop
2. The user makes a running-jump - how does the system know whether they're planning on sticking their landing, or continuing to run?
3. The user makes a running jump, but pivots in midair to run off at a direction different to the direction they jumped from

"mptp" wrote:
Here are three use cases that would cause anything but the most perfect sensing/engineering combination to sweep the user off their feet:
1. The user is sprinting forward at full speed, then slams their front foot downwards to come to an immediate stop
2. The user makes a running-jump - how does the system know whether they're planning on sticking their landing, or continuing to run?
3. The user makes a running jump, but pivots in midair to run off at a direction different to the direction they jumped from

Full body motion capture should be the most straight forward solution, and if you have this level of feedback you'd want it as an interface anyway. On it's own that doesn't help but using a kinematic model of humanoid locomotion it could anticipate possible outcomes for the next few milliseconds, and with 1khz sensors the latency could be in the sub-millisecond range.

"mptp" wrote:
I'd dearly love to be proven wrong, but it seems like the idea's been around for a while, and nobody quite knows how to solve the core issue of 'how do we control the direction and speed of the treadmill?'.

- 3rd generation omnidirectional treadmill : movement capture with eight high-definition cameras
- Torus Treadmill : automatic tracking with two Polhemus trackers
- CyberWalk : markerless motion tracking from multiple views

There've been numerous papers published about the implementation details of these devices. Virtual reality didn't appear last year, it's been a research subject for more than 50 years both in academics and military.

It's controlled a by human currently…

Mainly due to cost and non-portability of existing long-sidtance tracking systems.

In a technical sense, ODT and automated tracking systems are solved problems. You have the cash, you can go and buy one, and it will work. The actual problem is not building one, it's bringing the price down sufficiently to make it a mass-produced consumer device rather than a single-production research or military device.