SimToolReal: High-Fidelity AUV Simulation

• SimToolReal is a comprehensive simulation framework for multi-AUV missions that fuses advanced physical, environmental, and controller models for high-fidelity validation.
• It utilizes a modular architecture with plug-ins for hydrodynamics, bathymetry, and acoustic communication, achieving simulation speeds over 100 times real time.
• Rigorous validation through continuous integration, unit and integration tests, and field trial comparisons ensures reliable simulation-to-field performance.

SimToolReal refers to a simulation framework and methodology for accelerated, high-fidelity development and validation of multi-AUV (Autonomous Underwater Vehicle) missions, combining detailed physical/environmental modeling, real-time integration with deployed vehicle controllers, scalable performance, and rigorous validation across simulation and field testing. The system—centered around the LRAUV Sim stack—embodies a modular architecture capable of running significantly faster than real time (RTF > 100), supporting massive virtual experiments essential to uncovering subtle multi-robot failure modes prior to field deployment (Player et al., 2023).

1. Simulator Architecture and Dynamics Modeling

The core engine is implemented atop the modular Gazebo simulator, using DART for rigid-body physics. The architecture comprises a hierarchy of plugins:

• Hydrodynamics & Vehicle Dynamics: Employing a 6-DOF model based on Fossen’s equations for body-fixed dynamics, the system represents vehicle state with $\nu=[u,v,w,p,q,r]^T$ (body velocities/angular velocities) and $\eta=[x,y,z,\phi,\theta,\psi]^T$ (global pose/orientation). The equations capture inertia (including added mass), Coriolis/centripetal effects, linear/quadratic hydrodynamic drag, and restoring hydrostatic forces:

$$M \dot{\nu} + C(\nu) \nu + D(\nu) \nu + g(\eta) = \tau$$

All major actuation mechanisms—propellers (empirical $F_t = \rho D^4 K_T(J) n|n|$), fins ($F_l = 0.5 C_l \rho v^2 A$), and mass shifters—are explicitly modeled as modules.

• Bathymetry and Environmental Data: The bathymetry subsystem utilizes up to 1 m resolution tiled grids, loading only the tiles under each AUV’s footprint via level-of-detail streaming. Environmental 3D+time fields (current, temperature, chlorophyll) are indexed through per-axis red-black trees, with spatial and temporal queries achieving $O(\log n)$ via binary search and interpolation.
• Acoustic Communication: Plugins model acoustic modems (see next section) and allow explicit injection of communication constraints.
• Controller Bridge: An inter-process controller bridge synchronizes the real vehicle control loops and the simulated time, ensuring black-box hardware controllers can be validated in the loop.
• Data flow: At each physics step, the simulator publishes vehicle state, the controller consumes state and returns wrench/servo commands, actuator models compute forces, and the loop iterates.

2. High-Fidelity Acoustic Communication Modeling

Underwater communications are encoded via a hybrid physical/statistical model:

• Propagation Loss: Transmission loss ($\mathrm{TL}(d, f) = k\log_{10}(d) + \alpha(f)d$) captures frequency-dependent attenuation (per Thorp’s empirical formula) and spherical spreading ($k\approx 20$).
• Packet Handling: Received level ($\mathrm{RL} = \mathrm{SL} - \mathrm{TL}(d, f)$) is compared to a receiver threshold to induce a hard drop-off. Additional random bit-error induced packet losses can be superimposed.
• Latency and Bandwidth: Communication delays are set by geometric range ($d/c$, $\eta=[x,y,z,\phi,\theta,\psi]^T$0 m/s), and packets are serialized via explicit bandwidth capping (e.g., 1 kbps), introducing queueing delay for heavy traffic.

This abstraction supports accurate modeling of contention, delay, and physical link constraints, supporting mission concepts that hinge on intermittent/low-RF communications.

3. Faster-Than-Real-Time Execution and Performance Engineering

SimToolReal achieves a Real-Time Factor (RTF) far greater than unity:

$\eta=[x,y,z,\phi,\theta,\psi]^T$1

• Benchmarking: For a single AUV over a $\eta=[x,y,z,\phi,\theta,\psi]^T$2 mile world (high‐resolution bathymetry), with physics step $\eta=[x,y,z,\phi,\theta,\psi]^T$3 ms, RTF $\eta=[x,y,z,\phi,\theta,\psi]^T$4 100 is sustained on an 8-core AMD Ryzen 5900HX laptop (no GPU). Scaling to $\eta=[x,y,z,\phi,\theta,\psi]^T$5 vehicles is roughly $\eta=[x,y,z,\phi,\theta,\psi]^T$6 (e.g., $\eta=[x,y,z,\phi,\theta,\psi]^T$7; $\eta=[x,y,z,\phi,\theta,\psi]^T$8).
• Optimizations: The system leverages large, CI-tuned physics timesteps (20–30 ms), dynamic bathymetry tile streaming, multi-threaded plugin execution, and decoupling of graphics rendering from physics updates to maximize utilization.

This acceleration enables massive batch experiments for validation and regression testing, as well as exhaustive rare-event/fault injection sweeps not feasible in real time.

4. Validation: Continuous Integration, Numerical Testing, and Controller Integration

A rigorous CI-based regime underpins simulation trust:

• CI Pipeline: Every code push runs a full test battery on headless servers, sweeping across key physics step sizes.
• Unit Physics Tests: Validate buoyancy invariance, restoring moment oscillations, terminal drag, thruster/fin actuation laws, and mass shifter equilibrium directly against analytic/empirical reference results.
• Integration Tests: Real (black-box) vehicle controllers are evaluated in open and closed-loop trajectories, with outputs matched to mission XYZT path tolerances.
• Regression Checks: Mission-level performance and RTF stability are monitored at $\eta=[x,y,z,\phi,\theta,\psi]^T$9 ms settings.

This approach ensures physical fidelity, temporal stability, and controller compatibility for both synthetic and actual mission control software.

5. Deployment in Multi-Robot Field Experiments

A canonical field campaign demonstrates SimToolReal’s cycle:

• Mission Example: An “acoustic hot-bunking” scenario—maintaining continuous sampling of a drifting water mass by swapping a sampling vehicle (SV) with a relief vehicle (RV)—was prototyped fully in simulation. The protocol combines GPS-based waypoint closes, acoustic terminal homing (pure pursuit), and robust two-way acoustic handshake.
• Transfer to Sea Trials: In field tests, lessons regarding hydrodynamic model delta (e.g., homing speed increased from $$M \dot{\nu} + C(\nu) \nu + D(\nu) \nu + g(\eta) = \tau$$0 m/s for current authority), acoustic channel contention, and unmodeled mechanical artifacts (flexible fairings causing yaw oscillations) directly led to retuning and adaptation of both simulation and real vehicles. Simulation-to-field delta predominantly manifested in actuator time constants, which were retuned post-deployment based on in-water logs.
• Validation Outcome: Full mission criteria were met on the third at-sea trial, confirming real-world relevance of simulated failure modes and mission logic.

6. Systemic Significance and Future Directions

SimToolReal (LRAUV Sim) achieves a dual mandate: (i) providing sufficient dynamical and communication fidelity to surface subtle, emergent multi-robot failures, while (ii) delivering extreme performance (RTF > 100) for scaling up mission development and validation prior to expensive dock-to-sea transitions. The modular architecture—encompassing environmental streaming, multi-resolution environmental fields, explicit controller bridges, and acoustic network physics—facilitates both higher-level co-design (controls/behavior) and deep regression at the M&S level. A plausible implication is that such architectures will underpin broader classes of cooperative underwater and distributed-robot missions, enabling integration of more advanced real-in-the-loop or model-predictive autonomy in the future (Player et al., 2023).