pytorch / pytorch/rl

Feature Request: Expand ChatEnv with Popular Agent Environment Integrations

Open
#2,984 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
3.6k
Forks
484
Avg merge
1d 1h
Merged PRs (30d)
207

Description

The feature: Agent env integration

Building on top of torchrl.envs.llm.ChatEnv, we should expand the environment integrations to support a wider range of agent-tool interactions. This would enable training models that can interact with various tools (browsers, shells, games) through LLMs.

## Proposed envs

Here are the key environments and frameworks to consider integrating:

  • MLGym (facebookresearch/mlgym) for structured ML experimentation
  • BrowserGym for web interaction environments
  • MineDojo for rich Minecraft-based tasks with natural language specifications
  • AgentVerse for multi-agent interactions and tool usage
  • WebArena\
  • SWE-bench

The implementation would extend ChatEnv with a standardized interface for tool integration, consistent reward transformations, and unified observation spaces. This would greatly benefit research in tool-using LLM agents and provide standardized benchmarking across different environments.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with torchrl.envs.llm.ChatEnv and review how the checked MLGym integration is represented. Then compare the proposed BrowserGym, MineDojo, AgentVerse, WebArena, and SWE-bench integrations and define the scope of a standardized tool interface, reward transformations, and observation spaces. Done is not yet precisely defined beyond supporting the selected environments consistently.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.