Orchestrating Autonomous CCA Swarms
This challenge requires developers to build a decentralized multi-agent system for drone wingmen. You will design a hierarchy of autonomous agents capable of performing mission planning, sensor fusion, and tactical execution in a simulated contested environment. The core focus is on task decomposition: how a lead 'human-in-the-loop' agent delegates high-risk roles (e.g., electronic warfare, decoy, or kinetic strike) to autonomous wingmen while maintaining strict adherence to Rules of Engagement (ROE). Participants will utilize the AutoGen framework to manage agent conversations and the Qwen 2.5-72B model for tactical reasoning. The simulation must handle 'dynamic re-tasking'—where an agent must pivot its objective if a peer is neutralized or a new high-priority threat (like the Russian Oreshnik missile system) is detected. Success is measured by the swarm's ability to minimize attrition while achieving primary mission objectives within a defined physics-based simulation window.
What you are building
The core problem, expected build, and operating context for this challenge.
Implement a multi-agent orchestration layer using AutoGen and Qwen 2.5 to coordinate autonomous drone wingmen in complex tactical scenarios.
How work is evaluated
Evaluation is based on mission success rate, adherence to constraints, and the logical consistency of agent communication logs.
Shared data for this challenge
Review public datasets and any private uploads tied to your build.
What you should walk away with
Develop a robust 'Rules of Engagement' (ROE) validation layer to prevent unauthorized autonomous actions.
Design a hierarchical agent architecture for command and control (C2) using AutoGen.
Evaluate swarm performance using multi-objective optimization metrics (attrition vs. target success).
Implement real-time task allocation logic based on dynamic threat environments.
Integrate physics-based constraints into LLM reasoning to ensure aerodynamic feasibility.
Reference links and supporting material
Sample data for 1 tasks
How this agent runs
Evaluation is based on mission success rate, adherence to constraints, and the logical consistency of agent communication logs.
Challenge input
JSON defining mission parameters (Target coordinates, Threat locations, ROE).
Agent execution
The configured agent processes the input under the challenge policy.
Evaluated output
Log of agent communications and final state of assets.
- Ensures no kinetic action was taken before electronic warfare thresholds were met.
- Mission Success Rate target: 85
- Python execution harness
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- The evaluation module defines the checks.
Recipe state
This is a preview. The configuration can change before the evaluation recipe is locked.