Fixed reward overloading bug which results in incorrect rewards being used by POMDPs.solve

[ZhongxiaYan]

It looks like this reward overloading is causing issues with the SARSOP solver not outputting any meaningful solution. I was using this code on the RockSample task; when I looked at what reward is being written into the model.pomdpx, I see that all rewards are zero since both s and s' are always the same state. My environment is

macOS 11.2.2
Julia 1.3.1

(v1.3) pkg> st
Status ~/.julia/environments/v1.3/Project.toml
[8bb6e9a1] BeliefUpdaters v0.2.1
[159f3aea] Cairo v1.0.5
[a81c6b42] Compose v0.9.2
[a0b5b9ef] Cxx v0.4.0
[4b033969] DiscreteValueIteration v0.4.5
[7f35509c] POMDPGifs v0.1.0
[abefb91b] POMDPModelChecking v0.0.0
[08074719] POMDPModelTools v0.3.2
[182e52fb] POMDPPolicies v0.3.3
[e0d0a172] POMDPSimulators v0.3.9
[a93abf59] POMDPs v0.9.1
[92933f4c] ProgressMeter v1.4.0
[295af30f] Revise v3.1.12
[de008ff0] RockSample v0.1.3
[cef570c6] SARSOP v0.5.4
[f11abc24] Spot v0.1.1
[10745b16] Statistics

but I also had the same issue on a RedHat Linux server as well. Removing this reward line fixes the issue.

[MaximeBouton]

Thank you for pointing that out! I will take a look later today. I am not sure I understand why it failed with the current reward definition. I am wondering if it is due to any upstream changes (in POMDPModelTools or RockSample perhaps).

[MaximeBouton]

This should also be addressed by #26