// maze example (POMDP)
// slightly extends that presented in
// Littman, Cassandra and Kaelbling
// Learning policies for partially observable environments: Scaling up  
// Technical Report CS, Brown University
// gxn 29/01/16

// state space (value of variable "s")

//  0  1  2  3  4
//  5     6     7
//  8     9    10
// 11     13   12

// 13 is the target

pomdp

// can observe the walls and target
observables
	o
endobservables
// o=0 - observation in initial state
// o=1 - west and north walls (s0)
// o=2 - north and south ways (s1 and s3)
// o=3 - north wall (s2)
// o=4 - east and north way (s4)
// o=5 - east and west walls (s5, s6, s7, s8, s9 and s10)
// o=6 - east, west and south walls (s11 and s12)
// o=7 - the target (s13)
const double sl;
module maze

	s : [-1..13];
	o : [0..7];
	
	// initialisation
	[] s=-1 -> 1/13 : (s'=0) & (o'=1)
			 + 1/13 : (s'=1) & (o'=2)
			 + 1/13 : (s'=2) & (o'=3)
			 + 1/13 : (s'=3) & (o'=2)
			 + 1/13 : (s'=4) & (o'=4)
			 + 1/13 : (s'=5) & (o'=5)
			 + 1/13 : (s'=6) & (o'=5)
			 + 1/13 : (s'=7) & (o'=5)
			 + 1/13 : (s'=8) & (o'=5)
			 + 1/13 : (s'=9) & (o'=5)
			 + 1/13 : (s'=10) & (o'=5)
			 + 1/13 : (s'=11) & (o'=6)
			 + 1/13 : (s'=12) & (o'=6);
	
	// moving around the maze
	
	[east] s=0 -> (1-sl):(s'=1) & (o'=2) + sl:(s'=s) & (o'=o);
	[west] s=0 -> (s'=0);
	[north] s=0 -> (s'=0);
	[south] s=0 -> (1-sl):(s'=5) & (o'=5) + sl:(s'=s) & (o'=o);

	[east] s=1 -> (1-sl):(s'=2) & (o'=3) + sl:(s'=s) & (o'=o);
	[west] s=1 -> (1-sl):(s'=0) & (o'=1) + sl:(s'=s) & (o'=o);
	[north] s=1 -> (s'=1);
	[south] s=1 -> (s'=1);

	[east] s=2 -> (1-sl):(s'=3) & (o'=2) + sl:(s'=s) & (o'=o);
	[west] s=2 -> (1-sl):(s'=1) & (o'=2) + sl:(s'=s) & (o'=o);
	[north] s=2 -> (s'=2);
	[south] s=2 -> (1-sl):(s'=6) & (o'=5) + sl:(s'=s) & (o'=o);

	[east] s=3 -> (1-sl):(s'=4) & (o'=4) + sl:(s'=s) & (o'=o);
	[west] s=3 -> (1-sl):(s'=2) & (o'=3) + sl:(s'=s) & (o'=o);
	[north] s=3 -> (s'=3);
	[south] s=3 -> (s'=3);

	[east] s=4 -> (s'=4);
	[west] s=4 -> (1-sl):(s'=3) & (o'=2) + sl:(s'=s) & (o'=o);
	[north] s=4 -> (s'=4);
	[south] s=4 -> (1-sl):(s'=7) & (o'=5) + sl:(s'=s) & (o'=o);

	[east] s=5 -> (s'=5);
	[west] s=5 -> (s'=5);
	[north] s=5 -> (1-sl):(s'=0) & (o'=1) + sl:(s'=s) & (o'=o);
	[south] s=5 -> (1-sl):(s'=8) + sl:(s'=s) & (o'=o);

	[east] s=6 -> (s'=6);
	[west] s=6 -> (s'=6);
	[north] s=6 -> (1-sl):(s'=2) & (o'=3) + sl:(s'=s) & (o'=o);
	[south] s=6 -> (1-sl):(s'=9) + sl:(s'=s) & (o'=o);

	[east] s=7 -> (s'=7);
	[west] s=7 -> (s'=7);
	[north] s=7 -> (1-sl):(s'=4) & (o'=4) + sl:(s'=s) & (o'=o);
	[south] s=7 -> (1-sl):(s'=10) + sl:(s'=s) & (o'=o);

	[east] s=8 -> (s'=8);
	[west] s=8 -> (s'=8);
	[north] s=8 -> (1-sl):(s'=5) + sl:(s'=s) & (o'=o);
	[south] s=8 -> (1-sl):(s'=11) & (o'=6) + sl:(s'=s) & (o'=o);

	[east] s=9 -> (s'=9);
	[west] s=9 -> (s'=9);
	[north] s=9 -> (1-sl):(s'=6) + sl:(s'=s) & (o'=o);
	[south] s=9 -> (1-sl):(s'=13) & (o'=7) + sl:(s'=s) & (o'=o);

	[east] s=10 -> (s'=10);
	[west] s=10 -> (s'=10);
	[north] s=10 -> (1-sl):(s'=7) + sl:(s'=s) & (o'=o);
	[south] s=10 -> (1-sl):(s'=12) & (o'=6) + sl:(s'=s) & (o'=o);

	[east] s=11 -> (s'=11);
	[west] s=11 -> (s'=11);
	[north] s=11 -> (1-sl):(s'=8) & (o'=5) + sl:(s'=s) & (o'=o);
	[south] s=11 -> (s'=11);

	[east] s=12 -> (s'=12);
	[west] s=12 -> (s'=12);
	[north] s=12 -> (1-sl):(s'=10) & (o'=5) + sl:(s'=s) & (o'=o);
	[south] s=12 -> (s'=12);

	// loop when we reach the target
	[done] s=13 -> true;

endmodule

// reward structure (number of steps to reach the target)/7
rewards

	[east] true : 1/7;
	[west] true : 1/7;
	[north] true : 1/7;
	[south] true : 1/7;

endrewards

// target observation
label "goal" = o=7;
label "bad" = o=6;