Physical Residual Distillation for AUV Reinforcement Learning Control

Hybrid Stonefish-MuJoCo simulation and residual hydrodynamics distillation for AUV control.

**Dec. 2025 - Mar. 2026 Philadelphia, PA**

Developed a hybrid simulation workflow for autonomous underwater vehicle control that couples rigid-body dynamics with a learned residual hydrodynamics model.

Highlights

  • Connected Stonefish and MuJoCo to combine high-fidelity underwater simulation with efficient rigid-body simulation.
  • Distilled residual hydrodynamic effects into a learned model for use during reinforcement learning control.
  • Used the resulting simulation pipeline to support stable AUV control experiments.