PINN-Based Guidance for Air-to-Air Missiles

Physics-Informed Neural Network for real-time optimal trajectory generation, compared against classical guidance laws.

Documentation Index

Document	Description
Installation	Setup, dependencies, and verification
Physics	Coordinate systems, equations of motion, sign conventions
Simulator	Missile, Target, and Simulation components
Guidance	Classical guidance laws: PN, APN, AAPN, LQ
Optimization	CasADi-based trajectory optimization for training data
PINN	Neural network architecture, training, and inference
API Reference	Complete API documentation for all modules

Key Features

• High-fidelity simulator with ISA atmosphere, Mach-dependent aerodynamics (RASAERO/DATCOM), and AMRAAM-class missile parameters
• Multiple guidance laws: PN, APN, AAPN (energy-adaptive), and LQ optimal
• Trajectory optimizer achieving 100% convergence across 8 test scenarios using PN simulation for initial guesses
• Physics-informed neural network for real-time guidance inference

Research Objective

Apply PINN-based guidance to air-to-air missile engagement and compare computational efficiency and intercept performance against classical methods (PN, APN, LQ).

Quick Start

from pinn_guidance import Missile, MissileParams, Target, Simulation
from pinn_guidance import ProportionalNavigation

# Create engagement
params = MissileParams()  # AMRAAM-class defaults
sim = Simulation(
    missile_params=params,
    missile_x=0.0, missile_y=0.0, missile_z=-5000.0,
    missile_V=800.0, missile_gamma=0.0, missile_psi=0.0,
    target_x=10000.0, target_y=0.0, target_z=-5000.0,
    target_V=300.0, target_gamma=0.0, target_psi=3.14
)

# Run with PN guidance
guidance = ProportionalNavigation(N=4.0)
result = sim.run(guidance_func=guidance, max_time=30.0)
print(f"Intercept: {result['intercept']}, Time: {result['final_time']:.2f}s")

Recent Updates (2025-01-04)

• GPOPS-II Integration: Optimal trajectory generation using GPOPS-II + SNOPT solver
• Path Constraint Fix: Reduced from 4 to 2 constraints to fix active-set degeneracy (singular basis)
• SNOPT Configuration: Proper options via setup.nlp.options.snopt.xxx interface for R2019a compatibility
• Miss Distance: Achieved 10.00m (within tolerance) on test scenarios

Previous Updates (2025-12-23)

• AAPN Guidance Law: New energy-adaptive guidance with criticality-based N adjustment
• PN-based Initial Guess: Optimizer uses PN simulation for dynamically feasible starting points
• RASAERO/DATCOM Aerodynamics: Updated Mach-dependent CD0 and CN_alpha tables
• 8/8 Convergence: All test scenarios pass with CONSTRAINED optimization
• 8-second Motor Burn: Extended from 7s for better beam attack performance

See Implementation Summary for details.

Design Documents

See Design Document for full research scope and design decisions.