Skip to content

Latest commit

 

History

History
49 lines (31 loc) · 3.02 KB

EN.md

File metadata and controls

49 lines (31 loc) · 3.02 KB

Technical explanation

To implement our environments, we used the example in the gym_examples repository as a boilerplate. This decision facilitated the reuse of the graphical elements of a rendered environment using Pygame, allowing us to focus primarily on implementing the logic of each environment.

The constructed environments are:

PushBox

  • The simplest environment and the one upon which the others are based.
  • Its characteristics include:
    • In its initial state, an agent and a box or target are randomly placed.
    • The agent has four possible actions: move up, down, left, and right.
    • If the agent is in a position adjacent to the target and moves in the direction of the target, then it will move the target (i.e., the agent can push the box).
    • The box remains fixed in the environment unless the agent moves it.
    • The agent's objective is to push the box to a defined region of the environment.
    • The agent's policy is binary and sparse. This means that the agent will receive 0 as a reward in each step that it does not achieve the goal, and it will receive 1 when it does.

PushBoxRand

  • This environment has the same characteristics as PushBoxBaseEnv except for characteristic 4. In this case, the environment changes the box randomly every certain number of steps (defined by the size of the environment). Additionally, it adds a fifth action to characteristic 2: stay in the current position.

PushBoxRandPol1

  • It shares the characteristics of PushBoxRand but modifies the rewards that the agent receives in each step. This particular environment rewards movements with 2 points and penalizes staying in the same position with -1 point. It gives 100 points for fulfilling the objective.

PushBoxRandPol2

  • It shares the characteristics of PushBoxRand but modifies the rewards that the agent receives in each step. This particular environment penalizes movements with -1 point and rewards staying in the same position with 2 points. It gives 100 points for fulfilling the objective.

Getting started

For a complete overview of the project and how to use the environments, you can import the .ipynb file into Google Colaboratory. Local usage instructions can be found in the aforementioned file.

Additional notes

Due to the lack of a practical way to link commits from the README of a repository, the identifiers of each commit involved in the creation of each environment are provided below. This allows you to see what modifications were made from the original fork:

PushBox (Base Environment)

  • Updates to imports, modules, and module requirements: 03437e0
  • Graphic changes: 224e778, 58b5b5a, 2fee43f, 926fb90
  • Modification of the reset method logic: a0220b6, 5366444
  • Modification of the step method logic: 509e425, 5366444

PushBoxRand

  • Declaration in the module and creation of the environment: 10bb95a
  • Graphic changes: 7e50435
  • Change in the logic of the step method and in the initializer of the environment: e998be4, 88f92d5, d647df3