The lego nxt is an excellent platform to develop mobile agents. The robot I build is based on 2 light sensors in the bottom to detect the food colors and 2 contact sensors in the front to detect obstacles. From the bottom:

And from the top:

The first step is to calibrate the ligh sensors to adjust min-max and average value of the food patch in the playground. The robot does this automatically when it runs over it, then it starts its reaction behaviour to the food patch.

The agent learns the association of the grey area with the black area containing food and so it remains more time in the inner circle.

The controller's code is based on the nxtOsek firmware 2.07 and can be downloaded here.

The samples directory is organised in this way:

• nxtlif: an example that use a library for integrate and fire neurons
• nxtnav: a simple navigation library for odometry
• mybotslave and mybotmaster contains example for the bluetooth communication
• mybotsound: a controller using ICO learning
• mybotfull: the same controller but also using ultrasound for distance sensor

The leaky integrate and fire neuron is an important element in computational neuroscience: You can see the typical response to a stimulus:

There's a program you can use to log the bluetooth signals coming from the Lego NXT, it runs at the moment only on Windows XP, but I'm porting it also for Ubuntu.
The project for Visual Studio is available here.
In the control program you can decide what variables you want to log and they will be recorded on a CSV file and you can also plot them as you can see in this video.