Cognitive MFR Simulation Platform to Evaluate Radar Scheduling Algorithms

We present a novel simulation platform that advances multi-function radar (MFR) scheduling research by modeling complete radar-environment interactions. Our key innovation is reducing complex multi-sensor scheduling to a continuous online single-machine problem through task-sensor flexibility and simplified search functions, enabling development and evaluation of both traditional and learning-based scheduling approaches. The platform models full Singer target dynamics in 3D space, realistic radar physics, and multiple sensor interactions, achieving <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$10 \times$</tex> real-time performance while scaling linearly with available computational resources, in particular, the ability to simulate on GPUs. Implementation in JAX enables efficient parallel simulation of thousands of scenarios, facilitating development of reinforcement learning solutions. We illustrate use of the platform with Monte Carlo Tree Search (MCTS) and Proximal Policy Optimization (PPO) as well as simple scheduling heuristics across varying target densities and sensor configurations.