Using time in an influence map?

Community General Discussions
, , ,
1

I am working on some new AI for my bot ‘Eric’, and am using an influence map (actually mapS) to determine where interesting things are, such as shelter, light (solar), danger etc.

The problem is how to integrate time into my influence maps? I could maintain 24 different sets of maps but that seems very inefficient. Can anyone suggest how i can integrate time into the maps so that my bot ‘learns’ that sunlight is available in a certain location at a certain time, or even better that the light ‘influence’ moves over time.

I am coding this in c#, but that is kind of irrelevant as my bot uses a host of different platforms and languages, so its just some theory that i’m after. Thanks

2

Just a thought: create a

Just a thought: create a single nominal sunshine map (sun indexed by time) and add it as an influence, which may be ‘good’ or ‘bad’ depending on the influence with which it is interacting and itself may be influenced by other factors such as enclosure or cover.

3

Any reason you don’t

just add another dimension to an array that holds your location data to include time? Add an RTC to your robot to be able to timestamp your locations. If this is an outdoor bot, I would suggest a compass module. I am not sure how well a compass module would work inside.

4

agreed, another dimension to

agreed, another dimension to the array will be the best to record the location with the time and the data related to the location (shade, light, shelter, danger …)

 

Can you tell us more about “eric”, may be post some pictures ?

5

Thanks for the suggestions,
Thanks for the suggestions, a multi dimensional array should be easily implimented, so that could make a good start, possibly some interpolation between the times.

Eric is currently very simple, and changes on a daily basis, but he can be tracked or wheeled - trying to figur out whats better such as skid steer or real steering, i have a few basic sensors, such as ultrasonic, infra red, a microphone, and a light sensor. The breadboard started with a pic16f690, then a pic24@4mips, and now dual dspic33ep512s @ 70mips each (its a board i designed for work, but its VERY suited to this type of thing. The decision making can be done using an onboard raspberry pi or from the laptop using bluetooth. The pics are coded in c30, with the laptop using whatever gets the job done, c, c++, vb.net,c#, or python.

I will take some photos after work.

Bdk6 i like this idea, the only type of classification i have ever done was 2d classification in n-dimentional hyperspace, it was basically an algorithm that boiled down lots of parameters about an item into a single number, if the number matched you have a known item. This was calculated from 2d graphs using information such as the number of peaks, troughs, min,max, algeabraeic area under the curve etc, all boiled down. Do you know where i can find out more about how i could classify using the fourier transform of the data?

Influence maps, these are effectively an overlay on a map which store information about previous (as well as current) events, they are sometimes used in realtime strategy games to see where the ‘front’ is, and avoid areas where large casualties have happened, a chokepoint. I am trying to use them so that my bot can determine what to do and where to go. Fo example, if my bot falls down a small step, the bump will generate a spike on an acceleromter. This is bad, so the current location would have its danger influence increased relative to the acellerometer spike. The bot should then avoid that area unless some stronger ‘desire’ overrides it such as the need for ‘food’ existing only on the other side of the step.

I look forward to discussions with you all, just now im trying to learn my way about the site :slight_smile:

This was the screen for seeing the ping from the servo controlled ultrasonics. http://youtu.be/wdGDXjKDKpw

I am just starting to learn openGL, so this is what i have done on that front, this is testing of Erics ultrasonics, on a servo mount mapped into stationary 3d space http://youtu.be/7wQ0ZmecoeA

This was the initial test of the ultrasonic on the wheeled platform http://youtu.be/uWxOGkEwADU

The tracked platform http://youtu.be/kUREE-ceCL4

Influence maps http://aigamedev.com/open/tutorial/influence-map-mechanics/

Some rts stuff i dabbled with to see if i could do it http://youtu.be/clCbpEiUmC0 i have got working fog of war and queueable orders working now too

Some fun at work http://youtu.be/Y6swZbYHsAM