| Definition | A collection of gym default environments implemented with custom reward structures. |
| No of environments | 4 |
The custom_envs repository contains a set of environments built using the gym boilerplate, but diversifying them slighlty :
PushBox: The agent and a target (box) are placed randomly. The agent can move up, down, left, or right and can push the box if adjacent. The goal is to push the box to a defined region, with a reward of 1 for success and 0 for all other actions.
PushBoxRand: Similar to PushBox, but the target (box) changes position randomly after a set number of steps. The agent also gains an additional action to stay in place.
PushBoxRandPol1: Builds on PushBoxRand with a different reward structure, granting 2 points for moving and penalizing staying in place with -1. The agent earns 100 points for completing the task.
PushBoxRandPol2: Like PushBoxRand, but penalizes movement with -1 point and rewards staying in place with 2 points. The agent also earns 100 points for completing the task.
These environments provide varied challenges, testing agent adaptability and decision-making in dynamic settings.