Best practice for applying noise to sensors

Requests regarding how to set up experiments in ARGoS.
mbt925
Posts: 36
Joined: Mon Aug 07, 2017 12:21 pm

Best practice for applying noise to sensors

Postby mbt925 » Sat Mar 10, 2018 8:41 am

Greetings,

I wanted to apply noise to sensors reading and I was wondering if is there a formal way to determine how much noise and what type of noise should be applied to a sensor?

Thanks.

pincy
Site Admin
Posts: 294
Joined: Thu Mar 08, 2012 8:04 pm
Location: Boston, MA
Contact:

Re: Best practice for applying noise to sensors

Postby pincy » Sun Mar 11, 2018 9:11 am

I don't know of a formal way to do that. I typically try multiple levels of noise, to understand the role of noise in multi-robot coordination. The questions could be two, mainly:
  1. Is noise necessary? Sometimes it is, because it allows one to break symmetries
  2. How much noise is too much? It is often useful to check which level of noise breaks a certain algorithm, to test for robustness
I made ARGoS.

mbt925
Posts: 36
Joined: Mon Aug 07, 2017 12:21 pm

Re: Best practice for applying noise to sensors

Postby mbt925 » Sun Mar 11, 2018 1:32 pm

Thank you for the reply.
Usually, we apply noise to reduce the simulation-to-reality gap.
I am mainly concerned about how much noise and what type of noise will be there in the real-world robot sensors?

pincy
Site Admin
Posts: 294
Joined: Thu Mar 08, 2012 8:04 pm
Location: Boston, MA
Contact:

Re: Best practice for applying noise to sensors

Postby pincy » Sun Mar 11, 2018 6:39 pm

That depends on the robot and its sensors. It's typical to use Gaussian noise, but if you want a "perfect" simulation you need to take the real robot and derive the noise model of each sensor. There in no theoretical way to find realistic noise before you try the real robot.
I made ARGoS.


Return to “How to... ?”



cron