Kidnapped T265

[JimmyFroost]

|---------------------------------|------------------------------------------- |
| Camera Model | T265 |
| Operating System & Version | Windows 10 | |
| Platform | PC |
| Language | Matlab |

Hi

After experimenting T265 for a few months now, I've witnessed the phenomenom where the camera loses its way and finds it back.
Here is a small example where we see that the trajectory shifts a bit to the left and gets back to normal a few moments later :

Camera was set on a small cart, approximately 50cm above the ground.

My question is : if the camera is able to relocate itself , what can we do about false coordinates that we get during it's being kidnapped ?
The confidence is high (=3) but we still have bad positioning ?
Is there any way to fix the bad xyz that have been recorded ?

I cannot find any clear information about this on Intel's website , only a few general and vague stuff here

I read :

Our How-Tos and video tutorials cover everything from basics to advanced techniques.

but I can't find anything. Hopefully I am mistaken and someone can help me ?

I also have a D415, can the use of both of them improve this?

thank you

[ArtlyStyles]

how fast was the cart moving?

[JimmyFroost]

I was walking and pushing the cart quite slowly, slower than a usual walking speed.
I don't need to go fast for the project I'm working on, I'd rather say the slower the better

[JimmyFroost]

Another day, some other tests, some other kidnapping events !

Let me add this question :
How come the confidence value is still 3 (ie high) when the camera loses its way ? Is there no indicator that allows to understand:

• camera IMU is getting "moving state" value
• camera fisheyes see that new elements are coming in display, aka camera is moving

and yet the trajectory is still, or speed is almost null ?

[PaulKrush]

See issue #3516 and PR #4321

There are two different ways to look at this. Some people want the jump not to happen. I am like you and I want to pose history updated so the jump does not exist in the recorded path.

[radfordi]

Yes, I think you'll want to disable jumping once #4321 gets merged and the firmware gets updated.

[JimmyFroost]

Thank you @PaulKrush & @radfordi , I remember seeing this #4321 indeed.

Hopefully we'll have it in an update soon !

[RealSenseCustomerSupport]

---

The "pose jumping" option integrated in 2.25.0.

[JimmyFroost]

Sorry for "re-opening" this issue, but I'm not sure if this is possible with Matlab ?
I'm understanding it's not yet.

[radfordi]

Thanks for pointing out that the Matlab wrappers needed updating @JimmyFroost. Here's a PR to address that: #4666.

[JimmyFroost]

Thanks 👍