Skip to content

patronet-openenv

Patronet OpenEnv trains AI agents for critical emergencies like fires, wars, and disasters.

Code Demo Video Cerebral Valley

Core trick

GRPO trains a model over varied crisis scenarios to triage victims and dispatch responders within a 20-step limit.

Tags

agents

Description

Patronet OpenEnv trains AI agents for critical emergencies like fires, wars, and disasters. It simulates high-stakes scenarios, teaching AI to triage victims and dispatch the right responders within a strict 20-step decision window. Using OpenEnv and TRL’s GRPOTrainer, the system learns from diverse crisis paths, improving guidance, coordination, and communication. Packaged as a Docker FastAPI microservice with a visual simulation, it shows how AI decisions can save lives. Patronet OpenEnv equips governments and emergency agencies to respond faster and smarter when every second counts.
rkorrapolu/patronet-openenv★ 01 forksPythonCreated Mar 7, 2026 · last push Mar 8, 2026

README excerpt

---
title: Patronet Emergency Environment Server
emoji: 🚨
colorFrom: red
colorTo: blue
sdk: docker
pinned: false
app_port: 8000
base_path: /web
tags:
  - openenv
---

# Patronet OpenEnv

> An OpenEnv RL training environment for emergency response agents

This is a hackathon project that combines the OpenEnv specification, Berkeley RDI's AgentBeats evaluation paradigm, and Markov Decision Process (MDP) parameterized curriculum learning to build a reinforcement learning training environment for emergency response agents.

Current government emergency systems (911, 112, etc.) were designed for a pre-smartphone, pre-AI world. They assume you can speak clearly, someone will answer, and infrastructure is intact. In a real crisis, all three assumptions collapse simultaneously.

## Terminology

1. **OpenEnv** is an open source framework from Meta and HuggingFace that provides reusable environments for training and deploying AI agents. The framework uses Gymnasium style APIs `reset()`, `step(action)`, `state()`, and `close()` running as FastAPI servers inside docker containers. Each environment runs as a containerized microservice and integrate with GRPOTrainer through the `rollout_func` argument for rewards.

2. **AgentBeats** turns benchmarks into agents. The platform distinguishes between Green and Purple agents. Green agents are evaluator agents that define environments, issue tasks, collect results, compute metrics. Purple agents are competing agents under test. All agents commun

Team