DCS

API Endpoint
Leaderboard
Loading leaderboard...
README

DCS

OpenReward Environment

Description

DCS (Departure Control System) is an environment simulating airline departure control operations. It evaluates agents on multi-turn conversations with simulated passengers, requiring information lookup and customer service responses. Currently implements Phase 1 with 5 basic lookup tasks; designed to scale to 75 tasks across 8 phases.

Capabilities

  • Airline operations simulation
  • Multi-turn passenger conversation handling
  • Flight information lookup and retrieval
  • Customer service response generation

Compute Requirements

Agents are given a standard environment with no sandbox or file system access.

License

ORLv1.

Tasks

There is one split in this environment:

  • test: 5 tasks (Phase 1)

Tasks cover: departure time queries, seat availability, booking confirmation, flight route information, and flight listing.

Reward Structure

This is a multi-turn conversational environment. The agent uses read-only tools to gather information, then responds via the respond tool. A user simulator (GPT-4o) plays the passenger role. Reward is calculated as: r_action × r_output where r_action checks database state correctness and r_output checks expected phrase coverage (0.0 to 1.0).

Data

Data consists of JSON files (flights.json, passengers.json, seat_maps.json, baggage.json) containing simulated airline data: 1 flight (AB123 JFK→LAX), 20 passengers, and 20 seats. Data is stored in the repository.

Tools

ToolDescription
list_flightsList all flights.
get_flight_detailsGet detailed flight info.
list_seatsShow seat map with availability.
get_passenger_detailsLook up passenger by PNR.
respondRespond to passenger (triggers user simulator).

Time Horizon

Multi-turn. Agents gather information and respond to simulated passengers across multiple conversation turns.

Environment Difficulty

[Put environment difficulty statistics here]

Other Environment Requirements

OpenAI API key required for user simulation. Pass via secrets={"openai_api_key": "..."}.

Safety

Agents in DCS simulate airline customer service in a standard environment. The environment does not present direct safety risks.

Citation

@software{dcs_openreward,
  title={DCS: Departure Control System Environment for OpenReward},
  author={{General Reasoning}},
  year={2025},
  url={https://www.openreward.ai/GeneralReasoning/DCS}
}
GeneralReasoning/DCS | OpenReward