Basic AI motivation factors

[paulstreats]

Robots are generally made autonomous to do certain tasks, but these are usually very much predefined.

If you want to make a robot that is intelligent you need to make it able to create its own tasks to do, this is kind of where im up to, but the major problem to overcome seems to be giving some motivational needs factors to make the robot want / need to do tasks.

the following factors are what i have thought.

#power source (need to recharge when batteries become low or every so many hours)
#Get into signal range(need to update its program/memory every so many hours)

My list is pretty short as you can see, i was wondering if anybody had any more suggestions...

People have basic motivating needs such as food, sex etc... controlled by hormonal signals at certain times

from these basic needs other needs are built up such as financial(knowing that you have enough money to buy your food), security(knowing that nobody can pinch your food)

[JonHylands]

The main motivation other than the two you have listed should be to follow orders/goals given to it by its creator/owner.

Handling that introduces a huge can of worms, once you start thinking about the ramifications...

- Jon

[paulstreats]

the main reason behind my motivation system is that the user wont have to give any commands/goals, these should be created by the robot dynamically to help achieve the base motivation factors

[Steve Joblin]

Sounds like Abraham Maslow might be able to help... http://en.wikipedia.org/wiki/Maslow's_hierarchy_of_needs

[Admin]

Other ideas . . . avoid getting stepped on by people!!!

Perhaps looking for new hiding spots . . . or the goal of mapping/finding something. Or locating/chasing the cat around the house . . .

[Ro-Bot-X]

How about cleaning up? Or look for guests and if found, serve them beverages and cookies...

[paulstreats]

The actual mapping part of the robot will probably be always happening in the background, although im going to implement an inquisitive type of thing that can come into play when the robot gets bored such as exploring new places (maybe even gently nudging obstructions to see if they are moveable objects or not).
Maybe if the robot got some type of need satisfaction such as an energy boost when it serves cookies or greets guests, it may add into its learning structure.
Im currently working on a system of self propagating fuzzy tables, ive been working on and off this idea for 3 months now since i found out its not as easy as I first thought (the fuzzy tables are going to be used as a way of weighing up problems/setting and achieving targets but it seems to require a lot of depth such as nested tables and then youve got to decide on a nesting level. i seem to have come across a problem where sometimes a nested table may require a decision from one of its container tables to complete properly and it runs into infinite loops).  To start the process off, the super table will generally be set around basic needs and wants...

[Fredrik Andersson]

I've though about this concept a lot too. Never thought about a way to implement it though.

I think the motivation need to be really really simple, for now. An more advanced task requires a certain understanding about the situation and the more advanced the task is, the more information about it to solve it is needed and thus it would require a huge database, kind of like the memory in our brain. No... the motivation need to be simple, something that wouldn't need much understanding to grip. For example (this may sound like an pathetic approach but its just for researching purpose) being near a ball with a certain color or avuiding. The robot would start off with nothing more than it's motivation, meaning no code for tracing the ball. It should have code for getting position feedback from the camera or touch sensors if it bumps into something, only that the robot doesn't know what to do with the data. It tries something random out and compares that with the former values and so on...

Something else that i think is important in this concept is keeping the very lowest routines from the motivated thinking, like reflexes. The body and sensory is a whole other system, which communicates with the brain with a high level. The brain then perharps could override some functions and replace the behaviour of the low level functions if it works better that way.

Just some thoughts...

[Admin]

It tries something random out and compares that with the former values and so on...

Simple tasks (like drive forward) or medium tasks (figure out how to drive towards an orange ball) are doable. Its been done on robots many times.

The problem with this is sometimes the sequence of events is long or complicated to achieve a task, with no 'reward' until its done. And sometimes those 'values' are wasted memory or barely noticeable and therefore ignored. It needs to decide which values are salient because processing power isnt infinite.

And some tasks have a high penalty for failing (like driving). You can't do trail/error for these tasks (oh, water apparently shorts my battery! oops!). The robot needs an accurate internal future prediction algorithm, too.

But what about a robot learning how to build a robot, even with all the parts right in front of it? Too many steps . . .

[Asellith]

Just to add to this thread. Leaf is a windows based robot about that has some autonomous features to it. I think that system is based on a video games AI coding. It is programed in lisp and runs on windows. I have done a little research into the program but not much because I want something like that only in a linux environment. Because windows has way to much overhead and I don't want to deal with it right now. That is I already know windows. I don't know linux and should. Evidently from what I read the leaf robot includes roborealm software and animation software to produce an animated face that will lip sync to the text to speech algorithm in Leaf and leaf can understand verbal commands. It also has some facial recognition software in it. Neat little robot but I don't want to take the time to learn lisp so the system is a little confusing to me. Check it out at

http://tech.groups.yahoo.com/group/leaf_an_AI_robot

Johnny B