Even with the 320 KB of memory that the NXT supports, in order to implement both the heuristic and the adaptive algorithm for the robot, it was clear early in the design that the NXT’s memory would be insufficient for this program. However, with the program running entirely on the computer the Bluetooth link became a bottleneck. This can happen during any kind of motor movement when the rotation sensors are being queried to determine the amount of movement actually undertaken. This problem was addressed by distributing the processing, allowing part of the controller to run on the NXT and the other part to run on the computer. The standard NXT firmware and LEGO software (NXT-G) was used to program the brick to respond to Bluetooth commands, whilst the iCommand Java package, release 0.5, was used on the controlling computer in order to send and receive data through the Bluetooth link.

This probably doesn't contribute anything to the conversation, but I it's sort of relevant, so I'll say it anyway. I was programming a simple Q-learning algorithm a while back, I had this really weird moment when I kept thinking about robot ethics. I mean, in Q-learning, you're basically programming the robot with rewards and punishments. It's like giving it pain or pleasure depending on the action. (I wasn't inebriated, I promise).

Wow... you have interesting thoughts...
For sake of argument, though... (Even if it were an "ethical" question (I personally don't think robots are bound under the same moral code as humans).) People do the same thing to other people. How does a parent train a child? By rewarding them when they do right, and by punishing them when they do wrong. So, why would it be wrong to punish and reward a robot?

(This could get weird really quick)

The problem is the level of the punishment. All parents are confronted with this question.

[I hope my English is understandable ]