Next: The Simulated System Up: PhD thesis Previous: Future Work
In this chapter we describe the experiments we have carried out through simulation. We have used simulation for three different tasks: firstly, to check that the multiagent Navigation system we have designed works properly; secondly, we have applied Reinforcement Learning techniques in order to learn a policy on the use of the camera; and finally, we have used a Genetic Algorithm approach to tune the parameters of the agents in the Navigation system.
For these different tasks, we have used two simulators. We started
using the Webots![[*]](footnote.png)
simulator. On this simulator we implemented the Navigation system and
we also used it for the Reinforcement Learning task. However, we found
some problems with the Webots simulator, mainly related to batch
execution, which made the experimentation very slow. Although we were
able to get results when used for Reinforcement Learning, we decided to
develop our own simulator, to do extensive simulation with no
problems. We used this new simulator to run again the multiagent
Navigation system, and for the Genetic Algorithm approach to tune the parameters.