Observations

Observations#

In Flatland, you have full control over the observations that your agents will work with. Three observations are provided as starting point. However, you are encouraged to implement your own.

Provided Observations#

The three provided observations are:

  • Global grid observation

  • Local grid observation

  • Tree observation

stock observations

Global, local and tree: A visual summary of the three provided observations.

🔗 Provided observations

Code reference

The provided observations are defined in envs/observations.py

Each of the provided observations has its strengths and weaknesses. However, it is unlikely that you will be able to solve the problem by using any single one of them directly. Instead you will need to design your own observation, which can be a combination of the existing ones or which could be radically different.

Custom Observations#

🔗 Create your own observations

Gym Observation Builder#

There is GymObservationBuilder to take an existing ObservationBuilder and make it usable for RLlib (RayMultiAgentWrapper) and PettingZoo ( PettingZooParallelEnvWrapper):

  • FlattenedTreeObsForRailEnv: Gym-ified and flattened normalized tree observation.

  • GlobalObsForRailEnvGym: Gym-ified multi-agent GlobalObsForRailEnv.

This feature was introduced in 4.0.4

Observation Perturbations#

Perturbing observations can be implemented as wrapper around existing observations. The following two wrappers are implemented:

  • Wrapper to add Gaussian noise to numpy array based observations.

  • Tree obs wrapper to make trains blind for N timesteps according to Poisson.

This feature was introduced in 4.2.0

Contents