<!-- filepath: /home/philip/Documents/elsciRL-Wiki/Documentation/Instruction_Following/elsciRL_instruction_search.md --> # elsciRL Instruction Search This module provides the `elsciRLSearch` class, which enables unsupervised discovery of sub-goals and instruction mapping in RL environments. It is used to generate instruction paths and observed states for instruction-based RL experiments. ## Main Class: `elsciRLSearch` ### Features - Unsupervised exploration to discover sub-goals and instruction mappings - Supports multiple agent types and adapters - Produces observed states and instruction results for downstream instruction following - Integrates with elsciRL's environment and agent modules ### Key Methods - `__init__(...)`: Initializes the search with configs, environment, adapters, and exploration parameters - `search(action_cap=5, ...)`: Runs the exploration and instruction mapping process ## Usage Example See the example in the [Instruction Following documentation](./elsciRL_instruction_following.md) for a full workflow combining instruction search and instruction following. ---