Open Access Open Access  Restricted Access Subscription Access

ARTEMIS: A DEEP REINFORCEMENT LEARNING ARCHITECTURE FOR AUTONOMOUS COUNTER-UAS ENGAGEMENT WITH DETERMINISTIC FIRE CONTROL

Ashton J. Steele, Jean-Michel Dhainaut, Maddox Morrison, Dave Everett

Abstract


Small Unmanned Aircraft Systems (UAS) have emerged as a persistent, low-cost threat to ground forces, critical infrastructure, and mobile convoys. Most existing counter-UAS systems rely on fixed rule-based engagement logic that cannot adapt to changing sensor conditions, target behavior, or swarm tactics. This paper presents ARTEMIS (Autonomous Real-Time Engagement and Multi-Target Intercept System), a feasibility study of a simulation-validated architecture that combines a deterministic fire-control module with a Deep Reinforcement Learning (DRL) target-prioritization policy. The proposed framework integrates radar, optical, thermal, and laser-rangefinder sensor models with physics-based ballistics in a hardware-informed simulation environment and evaluates the learned policy using paired comparisons against deterministic baseline strategies under identical operating conditions. Single-drone validation achieved an 84.0% neutralization rate with 100% shot accuracy for all firecontrol- authorized engagements. Multi-drone validation across five scenario difficulty tiers demonstrated that, although the current policy underperforms the strongest deterministic baseline in overall kill rate, it consistently exhibits lower passthrough and timeout rates, indicating decisive target commitment. Analysis identifies the primary limitation as the current reward structure, which does not sufficiently encourage dynamic target reallocation under time-critical conditions. The modular architecture, which separates learned tactical decision making from deterministic weapon authorization, provides a safe and extensible framework for future improvements, including adaptive sensor trust weighting and reward refinement.


DOI
10.12783/ballistics26/37591

Full Text:

PDF

Refbacks

  • There are currently no refbacks.