Fishwood

Action Space	Discrete(2)
Observation Shape	(1,)
Observation High	[1]
Observation Low	[0]
Reward Shape	(2,)
Reward High	[1. 1.]
Reward Low	[0. 0.]
Import	mo_gymnasium.make("fishwood-v0")

Description

The FishWood environment is a simple MORL problem in which the agent controls a fisherman which can either fish or go collect wood. From Multi-objective Reinforcement Learning for the Expected Utility of the Return.

Observation Space

The observation space is a discrete space with two states:

• 0: fishing

• 1: in the woods

Action Space

The actions is a discrete space where:

• 0: go fishing

• 1: go collect wood

Reward Space

The reward is 2-dimensional:

• 0: +1 if agent is in the woods, with woodproba probability, and 0 otherwise

• 1: +1 if the agent is fishing, with fishproba probability, and 0 otherwise

Starting State

Agent starts in the woods

Termination

The episode ends after MAX_TS=200 steps

Arguments

• fishproba: probability of catching a fish when fishing

• woodproba: probability of collecting wood when in the woods

Credits

Code provided by Denis Steckelmacher