I'm currently looking at investigating swarm robustness in the present of noise, and would like to add footbot sensor/actuator noise. The sensor/actuators I'm interested in all have the ability to add noise, but the interface is very inconsistent:
In the battery, you specify the uniform range to draw noise from directly in the XML.
In the differential steering actuator, there are several variables used to calculate the actual applied
noise from a small equation.
In the differential steering sensor, you specify the uniform range to draw noise from for the velocity
and distance readings directly in the XML.
In the proximity sensor, ground sensor, light sensor, you just specify a noise level, which ARGoS
using internally to construct the range to draw noise from.
In the blob camera sensor, you specify the std deviation of the guassian distribution for the distance measure, but the angle measure is always drawn from a uniform distribution (0, 2*pi).
As you can see, there is quite a lot of variety in the level of control, as well as how things are specified in the XML. My thoughts on how to make things somewhat more consistent:
For the simpler sensors/actuators (battery, proximity, light), choosing whether to specify the range directly in the XML or as a single number is an easy change to make, and my preference is pythonic in this regard, in that explicit is better than implicit, and I think specifying the range directly in the XML is the way to go. For the blob sensor, I think exposing the angle of the noise in addition to distance to the user would also be good.
Some sensors/actuators allow the user to specify Gaussian distributions for noise (blob camera, differential steering), and some don't (proximity, ground, and light sensors). I think that adding the ability to choose the type of noise (i.e. uniform or Guassian), or making them all the same would be beneficial as well. However, I'm not a robotics expert, and there may be good reasons why some have uniform vs. Guassian depending on the type of sensor/actuator, but these are not documented in the argos3 -q <query> interface.
@ilpincy @allsey87 any thoughts how how to make things more consistent? I would be happy to do the implementation myself, if there is some consensus on what changes make sense and would enhance usability in ARGoS.
I'm currently looking at investigating swarm robustness in the present of noise, and would like to add footbot sensor/actuator noise. The sensor/actuators I'm interested in all have the ability to add noise, but the interface is very inconsistent:
As you can see, there is quite a lot of variety in the level of control, as well as how things are specified in the XML. My thoughts on how to make things somewhat more consistent:
For the simpler sensors/actuators (battery, proximity, light), choosing whether to specify the range directly in the XML or as a single number is an easy change to make, and my preference is pythonic in this regard, in that explicit is better than implicit, and I think specifying the range directly in the XML is the way to go. For the blob sensor, I think exposing the angle of the noise in addition to distance to the user would also be good.
Some sensors/actuators allow the user to specify Gaussian distributions for noise (blob camera, differential steering), and some don't (proximity, ground, and light sensors). I think that adding the ability to choose the type of noise (i.e. uniform or Guassian), or making them all the same would be beneficial as well. However, I'm not a robotics expert, and there may be good reasons why some have uniform vs. Guassian depending on the type of sensor/actuator, but these are not documented in the
argos3 -q <query>interface.@ilpincy @allsey87 any thoughts how how to make things more consistent? I would be happy to do the implementation myself, if there is some consensus on what changes make sense and would enhance usability in ARGoS.