maestro is an attempt at a naive sensorimotor management. how far can we get without deep learning or other advanced ai techniques?
step 1a. get a state representation abd -> [0,1,3] 110100 step 1b. add on your last action and encode it into an SDR.
step 2. keep a running list of the last thing we've seen. pass that up. abc, abd -> 
step 3. keep track of what leads to what in SDR terms
abc -> abd
[0,1,2] -> [0,1,3]
111000 -> 110100
step 4. get a goal from above, given where we are translate that into actions [0,1,7] -> a b e 110010 -> ^ ^ ^-state | |---state |-----suggested action, if there is no other way to get to be state then use the other action. consult where you are now 110100 -> 111000 abc 110100 -> 110010 abe <-- our goal and our action is a to get there. 110100 -> 010101 gbd
So the data hierarchy might look like this (each letter is a state (with acts)) histories | current moment -> goal o p n i m e b y | l u v e h i j k -> g o a l m e t o l u v e | h i j k g o a l h i | j k g o j | k g
in the next time step... histories | current moment -> goal p n i m e b y l | u v e h i j k g -> o a l m e t o g u v e h i j k g o a l g i j k g o a k g o