When the agent is not on a rock, it can sample, but it shouldn't be allowed.
A wrong action index causes that in actions.jl at line 2.
const BASIC_ACTIONS_DICT = Dict(
:north => 1,
:east => 2,
:south => 3,
:west => 4,
:sample => 5)
function POMDPs.actions(pomdp::RockSamplePOMDP{K}, s::RSState) where K
if in(s.pos, pomdp.rocks_positions) # slow? pomdp.rock_pos is a vec
return actions(pomdp)
else
# sample not available
return 2:N_BASIC_ACTIONS+K
end
end
as you can see if the agent is on a rock the function returns all the actions except the north action but the comment says it should return all actions except the sample one.
When the agent is not on a rock, it can sample, but it shouldn't be allowed. A wrong action index causes that in actions.jl at line 2.
as you can see if the agent is on a rock the function returns all the actions except the north action but the comment says it should return all actions except the sample one.