3 ms·

That and the fact that the PSVR2 uses camera-based SLAM tracking, similar to the Oculus Quest, and reimplementing that from scratch is non-trivial to say the le

That and the fact that the PSVR2 uses camera-based SLAM tracking, similar to the Oculus Quest, and reimplementing that from scratch is non-trivial to say the least.

ChuckNorris89 4y ago
Isn't the SLAM tracking processing done on the headset itself and then the computed movement data sent to the PC/PS5 via USB?
- jsheard 4y ago
  Is it? I haven't seen it definitively confirmed either way, but intuitively it would make more sense to do SLAM on the PS5s main processor rather than driving up the PSVR2s BOM with an on-board processor fast enough to do it. They have to stream the camera frames back to the PS5 anyway for the passthrough feature to work.
  - ChuckNorris89 4y ago
    >driving up the PSVR2s BOM with an on-board processor fast enough to do it The Quest 2 can do SLAM on the head set plus the 2*2K resolution 3D rendering in real time on a 4 year old QUALCOMM chip running Android OS on top and it costs less than the PSVR2. So Sony can definitely afford to put a processor in the headset just for the SLAM compute alone since the 3D rendering is done on the PS5. The chip in any optical mouse is a low resolution camera and DSP computing SLAM algos at hundreds of times a second and costs peanuts. VR headsets have more resolution to process and account for the third dimension in space but they also don't have to process the entire picture but just the differential movement of the LEDs captured in their blanking intervals, processed as white dots in B/W images, further simplifying things. So pretty sure the PSVR2 SLAM compute can be done on an ARM chip or FPGA worth ~ 10 USD nowadays.
makomk 4y ago
There's already a mostly-working open source SLAM implementation that people have been using on a few of the existing headsets. What complicates matters here is that at least feature extraction seems to run on the PSVR2 headset itself so the existing code can't just be dropped in as-is. (Also, whilst the controllers work in basically the same way as many other VR controllers using a constellation of IR LEDs, there's not really good open source support for any of those controllers either.)

That and the fact that the PSVR2 uses camera-based SLAM tracking, similar to the Oculus Quest, and reimplementing that from scratch is non-trivial to say the least.

ChuckNorris89 4y ago

Isn't the SLAM tracking processing done on the headset itself and then the computed movement data sent to the PC/PS5 via USB?

jsheard 4y ago
Is it? I haven't seen it definitively confirmed either way, but intuitively it would make more sense to do SLAM on the PS5s main processor rather than driving up the PSVR2s BOM with an on-board processor fast enough to do it. They have to stream the camera frames back to the PS5 anyway for the passthrough feature to work.
- ChuckNorris89 4y ago
  >driving up the PSVR2s BOM with an on-board processor fast enough to do it The Quest 2 can do SLAM on the head set plus the 2*2K resolution 3D rendering in real time on a 4 year old QUALCOMM chip running Android OS on top and it costs less than the PSVR2. So Sony can definitely afford to put a processor in the headset just for the SLAM compute alone since the 3D rendering is done on the PS5. The chip in any optical mouse is a low resolution camera and DSP computing SLAM algos at hundreds of times a second and costs peanuts. VR headsets have more resolution to process and account for the third dimension in space but they also don't have to process the entire picture but just the differential movement of the LEDs captured in their blanking intervals, processed as white dots in B/W images, further simplifying things. So pretty sure the PSVR2 SLAM compute can be done on an ARM chip or FPGA worth ~ 10 USD nowadays.

makomk 4y ago

There's already a mostly-working open source SLAM implementation that people have been using on a few of the existing headsets. What complicates matters here is that at least feature extraction seems to run on the PSVR2 headset itself so the existing code can't just be dropped in as-is. (Also, whilst the controllers work in basically the same way as many other VR controllers using a constellation of IR LEDs, there's not really good open source support for any of those controllers either.)