ElonCoding's picture
Update README.md
011868a verified
|
Raw History Blame Contribute Delete
4.59 kB
metadata
title: Airport Flight Scheduling OpenEnv
emoji: ✈️
colorFrom: blue
colorTo: indigo
sdk: docker
sdk_version: 1.0.0
python_version: '3.11'
app_file: app.py
pinned: false

Airport Flight Scheduling Environment

OpenEnv submission — Real-world AI agent environment for airport operations

OpenEnv HF Space


Overview

The Airport Flight Scheduling Environment simulates the real-world task of Air Traffic Control (ATC) and airport operations management. An AI agent must:

  • Sequence arriving flights onto available runways
  • Respect wake turbulence separation rules (ICAO standards)
  • Assign appropriate gates (domestic vs. international)
  • Handle weather disruptions and runway closures
  • Manage fuel emergencies with priority overrides
  • Minimize total delay minutes across the episode

This is a genuine, high-stakes logistics problem. Airlines and airports spend billions optimizing these exact decisions.


Observation Space

Field Type Description
clock float Simulated minutes elapsed since episode start
flights List[Flight] All flights with ETA, weight class, fuel state, status
runways List[Runway] Availability, current occupant, available-at time
gates List[Gate] Occupancy, terminal, domestic/international flag
weather Weather Visibility, wind, active NOTAMs
recent_events List[str] Last 6 notable events for context
score_so_far float Cumulative reward

Action Space

Action Parameters Description
assign_runway flight_id, runway_id Clear flight to land on runway
assign_gate flight_id, gate_id Assign gate post-landing
delay_flight flight_id, delay_minutes Push ETA back by N minutes
divert_flight flight_id Send to alternate airport
clear_runway flight_id, runway_id Mark runway clear after use
declare_emergency flight_id Grant priority to fuel-critical flight

Tasks

Task 1: Single Runway Landing Sequence (Easy)

  • Flights: 5 arriving aircraft
  • Runways: 1
  • Challenge: Order flights correctly by weight class to minimize separation penalties
  • Expected score: 0.70–0.90

Task 2: Multi-Runway Gate Assignment (Medium)

  • Flights: 10 mixed arrivals
  • Runways: 2 | Gates: 8 (4 domestic, 4 international)
  • Challenge: Match gate types, balance runway load, handle simultaneous arrivals
  • Expected score: 0.50–0.75

Task 3: Storm Disruption Recovery (Hard)

  • Flights: 15 flights including 4 with low fuel
  • Event: Runway 2 closes at t=20 due to storm
  • Challenge: Reassign all R2 flights, handle emergencies, minimize diversions
  • Expected score: 0.30–0.60

Reward Function

Per-step:   −0.1 × count(overdue unhandled flights)
Landing:    +1.0 per safe landing, +0.3 bonus if on-time (<2 min delay)
Gate:       +0.5 per gate assigned, +0.5 bonus if within 5 min of landing
Conflict:   −5.0 per runway incursion or gate double-booking
Emergency:  +3.0 per fuel emergency resolved before critical
Diversion:  −2.0 avoidable diversion | −0.5 unavoidable
Episode:    Up to +10.0 based on throughput × on-time ratio − conflicts

Setup & Usage

Local Development

git clone <your-repo>
cd airport-env
pip install -e .

# Start server
uvicorn server.app:app --host 0.0.0.0 --port 7860

# Run baseline
export API_BASE_URL=https://router.huggingface.co/v1
export MODEL_NAME=Qwen/Qwen2.5-72B-Instruct
export HF_TOKEN=hf_your_token
python inference.py

Docker

docker build -t airport-env .
docker run -p 7860:7860 \
  -e API_BASE_URL=$API_BASE_URL \
  -e MODEL_NAME=$MODEL_NAME \
  -e HF_TOKEN=$HF_TOKEN \
  airport-env

API Endpoints

POST /reset   {"task_id": "single_runway_landing", "seed": 42}
POST /step    {"action": {...}, "task_id": "..."}
GET  /state   ?task_id=...
GET  /grade   ?task_id=...
GET  /tasks
GET  /health

Baseline Scores

Task Agent Score
single_runway_landing Qwen2.5-72B 0.74
multi_runway_gate_assignment Qwen2.5-72B 0.58
storm_disruption_recovery Qwen2.5-72B 0.41

Validation

openenv validate

All checks should pass: typed models, step/reset/state endpoints, task graders, openenv.yaml.