search

gongziyida / Infotaxis

An implementation of infotaxis based on the paper "‘Infotaxis’ as a strategy for searching without gradients"
MIT License
11 stars 2 forks source link

a question #2

Open svtary opened 3 years ago

svtary commented 3 years ago

hi , thanks for your guidance about infotaxis,i have a question after reading your code, in your code we have already known the position of the source ,actually we set up the (x,y)as the source position ,so it's resonable to calculate the get_hit_rate which needs the source position ,however ,when we are not in simulation environment , it is impossible to know the source position ,so it's hard to calculate the hit_rate and the possibility . i'm very confused about it,could u give me some explanation?thank u very much

gongziyida commented 1 year ago

Sorry for the very belated reply. You probably have figured it out, but just in case you didn't and move on: In infotaxis the agent has no knowledge about the actual location of the source but performs a Bayesian inference to calculate the posterior of each location being a source. An intuitive explanation is that the agent keeps track of a map indicating where it is more likely to find the source, and updates it during navigation. You can visualize this with my implementation.

Knowing the location of the source is not required at all as long as you have the environment model to deliver stimuli to the agent (it could be a black box).