Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Human Engagement with Realistic AI Combat Pilots

Ardian Selmonaj, Giacomo Del Rio, Adrian Schneider, Alessandro Antonucci

TL;DR

This work introduces a modular framework to enable real-time human–AI interaction in realistic air-combat simulations by training MARL agents in a JSBSim-based 3D environment and deploying them into VR-Forces via the DIS interface. The MARL model is framed as a Markov game and trained with MAPPO under CTDE, curriculum learning, and self-play, with three control policies (Attack, Engage, Defend) and a commander policy to select among them, achieving up to 80% wins in 10v10 scenarios when enemies share the policies. Real-time interaction is facilitated through OpenDIS, enabling mixed human-AI teams in VR-F where agents can either have their dynamics simulated locally or have actions driven by VR-F, supported by a dedicated training loop in JSBSim at 100 Hz. The work highlights ethical considerations, emphasizes human oversight, and outlines future directions toward imitation learning, hybrid algorithms, and broader DIS-enabled integrations for richer, safer defense training.

Abstract

We present a system that enables real-time interaction between human users and agents trained to control fighter jets in simulated 3D air combat scenarios. The agents are trained in a dedicated environment using Multi-Agent Reinforcement Learning. A communication link is developed to allow seamless deployment of trained agents into VR-Forces, a widely used defense simulation tool for realistic tactical scenarios. This integration allows mixed simulations where human-controlled entities engage with intelligent agents exhibiting distinct combat behaviors. Our interaction model creates new opportunities for human-agent teaming, immersive training, and the exploration of innovative tactics in defense contexts.

Towards Human Engagement with Realistic AI Combat Pilots

TL;DR

This work introduces a modular framework to enable real-time human–AI interaction in realistic air-combat simulations by training MARL agents in a JSBSim-based 3D environment and deploying them into VR-Forces via the DIS interface. The MARL model is framed as a Markov game and trained with MAPPO under CTDE, curriculum learning, and self-play, with three control policies (Attack, Engage, Defend) and a commander policy to select among them, achieving up to 80% wins in 10v10 scenarios when enemies share the policies. Real-time interaction is facilitated through OpenDIS, enabling mixed human-AI teams in VR-F where agents can either have their dynamics simulated locally or have actions driven by VR-F, supported by a dedicated training loop in JSBSim at 100 Hz. The work highlights ethical considerations, emphasizes human oversight, and outlines future directions toward imitation learning, hybrid algorithms, and broader DIS-enabled integrations for richer, safer defense training.

Abstract

We present a system that enables real-time interaction between human users and agents trained to control fighter jets in simulated 3D air combat scenarios. The agents are trained in a dedicated environment using Multi-Agent Reinforcement Learning. A communication link is developed to allow seamless deployment of trained agents into VR-Forces, a widely used defense simulation tool for realistic tactical scenarios. This integration allows mixed simulations where human-controlled entities engage with intelligent agents exhibiting distinct combat behaviors. Our interaction model creates new opportunities for human-agent teaming, immersive training, and the exploration of innovative tactics in defense contexts.

Paper Structure

This paper contains 4 sections, 2 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Reinforcement Learning Agent Model
  3. Interaction Process
  4. Conclusions and Future Work

Figures (2)

  • Figure 1: Scenes of VR-Forces as (a) First Person View (FPV), (b) combat situation, and (c) simulator cockpit.
  • Figure 2: In training, our custom environment (Env) simulates (sim) dynamics. Agents observe (obs) states and respond with actions (act). In deployment, DIS communicates (com) to VR-F either JSBSim states (dashed) or actions for state simulation. Human users interact with VR-F via input devices.