Hacker News Reader

Comment by ano-ther | Hacker News Reader

Here is a video of it in action: https://www.youtube.com/watch?v=owSJK7vMSLk

Interestingly, humans come very close to it during operation.

Not much different than everyday on sites everywhere with boom cranes. Which are arguably much more dangerous due to being less dexterous and no video. Notably 1:23 where it's similar to working with a crane.

That looks incredibly cool, and also instantly nausea inducing - I can’t imagine the video feed and the head gyros are fast enough to avoid that little bit of lag and judder that turn VR into a vomit machine.

The feedback on the arm is very cool, though.

FPV drone hardware has commercialized very cheap very low latency video feeds to headsets.

Not with headtracking though. Not sure that's reasonably possible.

FPV headsets can have head tracking. I don’t know cost now, but it was possible even in 2012/2013/2014 when I worked in that industry.

Huh, so there is, with physical servos moving a camera.

Any idea what latency numbers might look like? Seems unlikely the physical movement can be especially low latency.

What's the advantages of head tracking though? I don't see it's in anyway better than just joystick to control robot head movement.

It's a completely natural movement for the operator, that will require almost no practice or training, and also leaves the operator's hands free to operate other controls.

You’re already using two joysticks with your thumbs or thumbs and index to fly.

Why not? There's no need for the neck to have more than 1~5 ms of lag, and about 5~10 ms photon lag even with a framebuffer involved (if you align the rolling shutter of the camera to the row-by-row multiplexed display, say OLED @ 240Hz, you should manage 4~5 ms).

I wonder if it wouldn't be easier / more comfortable for the worker to just get a low-poly reconstructed 3D rendering of the environment. That way at least they could move their head as much as they want and still get a read-time feed in their goggle without the usual VR tricks.

Because said low res rendering will remove very important cues that workers repairing stuff need to see, like potential things cracks in materials.

just over so it on the cameras so there isn't a latency because you're not actually moving the giant robot head

Responses so far say to optimize neck latency or use a joystick but why not use a 360 camera and handle it in reprojection. First pass could literally be a gopro or insta360 streaming to the headset and you could eventually integrate cameras and do the stitching yourself. The tech already exists and works, and it removes moving parts and actuators

> that little bit of lag and judder that turn VR into a vomit machine.

This doesn’t apply to everyone, so it’s just a matter of training people with more tolerance.

I like how the robot head motion is linked to the operator's head motion, such that the robot is able to nod its head near 2:15. I also like the feedback control near 0:35.

now start collecting user movement data with the visual recordings and we can eventually automate all the things

Reply on news.ycombinator.com