Geometry-Aware Surrogate for Real-Time Hydrodynamics Estimation of Autonomous Ground Vehicles in Amphibious Environments
Ammar Waheed, Luke Gallantree, Zohaib Hasnain* · 2026
Read Geometry-Aware Surrogate for Real-Time Hydrodynamics Estimation of Autonomous Ground Vehicles in Amphibious Environments by Ammar Waheed, Luke Gallantree, Zohaib Hasnain* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-18543.txt.
Introduction
Ammar Waheed and two co-authors present, in a 2026 paper, a per-surface neural network surrogate for real-time hydrodynamics estimation of autonomous ground vehicles in water. Trained only on high-fidelity CFD data from two vehicles, it predicts geometry-resolved forces at under 0.9 milliseconds per sample. Held-out tests give longitudinal-force error of 13 percent and vertical-force error of 3 to 12 percent. Water wading trials confirm predicted drag follows quadratic speed scaling and buoyancy intercepts scale linearly with depth.
Catalog record
| CREATOR | Ammar Waheed, Luke Gallantree, Zohaib Hasnain* |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70575 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.18543
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Geometry-Aware Surrogate for Real-Time Hydrodynamics Estimation of Autonomous Ground Vehicles in Amphibious Environments Ammar Waheed, Luke Gallantree, Zohaib Hasnain* Abstract Autonomous ground vehicles operating in shallow water or flood-prone terrains require dynamic models that account for hydrodynamic forces. However, the simulation and planning tools currently available either lack the physical fidelity or are too computationally expensive to run in real time. This work presents a per-surface neural network surrogate that bridges this gap by predicting geometry-resolved hydrodynamic forces at real-time rates, trained entirely on high-fidelity CFD data from two geometrically distinct vehicles. A vehicle specific Signed Distance Field (SDF) provides per-surface submergence inputs, allowing the model to resolve how loading varies with vehicle geometry, depth, and flow direction. On held-out CFD data, the surrogate achieves a longitudinal-force symmetric MAPE (sMAPE) of 13% and a vertical-force sMAPE of 3-12%, with inference running under 0.9 ms per sample. To evaluate the model under real-world conditions, water wading trials of a full-scale vehicle at different submersion depths are used. Motion capture derived kinematics serve as the surrogate inputs, and the resulting predictions are tested to reproduce known physical relationships between force, speed, and depth. The predicted drag follows quadratic speed scaling ( R 2 ≥ 0.97 R^{2}\geq 0.97 ) and the buoyancy intercepts scale linearly with depth ( R 2 = 0.973 R^{2}=0.973 ). Neither relationship is encoded in the model training loss, both emerge from the per-surface architecture summing individually predicted surface forces. The resulting framework provides a pathway for embedding physically grounded hydrodynamics into the simulation and planning loops that autonomous ground vehicles depend on in amphibious environments. I Introduction Accurate simulation is increasingly central to the development and testing of autonomous ground vehicles. For scenarios involving environmental complexity or safety-critical conditions, physical testing alone is often impractical due to high costs, limited repeatability, and risk to hardware [ 9 ] . This is especially true in fluid-affected environments such as shallow water or flood-prone terrain. The ability to predict how a vehicle will behave in such settings requires simulation tools that capture not just surface-level appearance, but also the underlying physics governing vehicle-environment interaction. Prior work has shown that even advanced simulators can exhibit significant sim-to-real gaps when key physical effects are omitted or overly simplified [ 6 , 36 ] . One critical yet under-represented interaction in many vehicle simulation frameworks is the dynamic influence of surrounding fluids on partially submerged systems. As ground vehicles enter shallow water or saturated terrain, hydrodynamic forces including drag, buoyancy, and added mass begin to dominate the dynamics [ 12 ] . These forces are not merely environmental perturbations; they alter the effective inertia, weight distribution, and net resistive forces experienced by the vehicle [ 2 , 22 ] . The interplay between vehicle geometry, velocity, submersion depth, and local flow state gives rise to highly nonlinear, coupled fluid-structure interactions that fundamentally change how the vehicle accelerates, turns, or maintains stability [ 25 ] . Such interactions cannot be neglected in applications ranging from unmanned amphibious mobility to flood-zone navigation, where decisions made by autonomous agents rely on underlying dynamic models. If the simulator fails to capture effects such as the transient drag increase when a platform pitches forward into water [ 17 , 18 ] or the lift-induced reduction in tire normal force [ 14 ] , the resulting models will be incomplete. As a result, controllers or planners developed in such a simulator may perform poorly or fail when applied in real-world conditions. Traditional solutions for modeling these coupled dynamics rely on Computational Fluid Dynamics (CFD) solvers, which can resolve pressure distributions, flow separation, and time-dependent hydrodynamic loads with high fidelity [ 30 ] . However, these methods are not feasible for interactive or real-time applications. Even simplified Navier-Stokes solvers with moving boundary conditions require extensive spatial discretizations and small time steps, often running several orders of magnitude slower than real-time, and full 3D simulations of vehicles transitioning between air, water, and ground modes can require hundreds of CPU-core hours to simulate seconds of physical time. At the other extreme, many robotics simulators and game engines approximate fluid effects with fixed drag coefficients or simplified buoyancy primitives, achieving interactive frame rates but without computing loads from the underlying flow physics [ 1 , 8 , 35 , 7 , 28 ] . These models do not resolve how forces vary with vehicle geometry, submersion state, or flow direction, and consequently the dynamics they produce do not transfer reliably to real hardware. The result is a persistent gap; the tools that are physically accurate are too slow for real-time use, and the tools that are fast enough lack the fidelity needed for meaningful sim-to-real transfer. This work addresses the aforementioned gap through a data-driven surrogate model trained on high-fidelity CFD data that predicts spatially distributed hydrodynamic forces across the vehicle body at real-time rates. The surrogate operates at the level of individual surface regions rather than producing a single lumped force, allowing it to resolve how loading changes with geometry and submersion state. The model is trained on two geometrically distinct platforms and validated experimentally on a full-scale vehicle at multiple submersion depths using only the measured kinematics as input. The experimental results show that the surrogate reproduces established hydrodynamic scaling laws that are absent from the training objective. By making geometry-aware force estimates available at interactive rates, this approach enables the kind of physically grounded hydrodynamic modeling that simulation and planning pipelines for autonomous ground vehicles currently lack. II Literature Review Understanding the behavior of vehicles in shallow water or flood‑like conditions requires modeling hydrodynamic forces that are absent on dry terrain. When a vehicle is partially submerged, buoyancy, drag, lift, and added mass become prominent and alter stability and mobility. Several studies focusing on amphibious and flood-driven vehicles have highlighted the importance of these forces for both mobility and stability [ 23 , 33 , 16 ] . Experimental and numerical investigations have defined critical thresholds for ”incipient motion”, where vehicles begin to float or slide under modest water flow, leading to instability [ 5 ] . High-fidelity studies in the vehicle hydrodynamics literature solve the incompressible Navier-Stokes equations with free-surface tracking to resolve pressure distributions and unsteady forces on vehicle bodies. RANS-based workflows have been used to quantify amphibious-vehicle resistance [ 19 ] , and form drag has been shown to account for 40-80% of total resistance at low planing speeds [ 25 ] . Tightly coupled multibody-CFD simulations have further demonstrated the interdependence of terrain mechanics and coastal hydrodynamics during land-to-water transitions [ 34 ] . However, the fluid-structure coupling is nonlinear [ 15 ] and depends on immersion level, speed, body geometry, and transient motions. The diversity of vehicle platforms, from compact wheeled robots to large tracked carriers, introduces further variation in wave patterns and added-mass effects [ 24 ] , so that no single analytical model covers all geometries and operating regimes. Consequently, accurate force prediction requires either high-fidelity simulation or calibration from empirical data. II-A Computational Fluid Dynamics (CFD) and Fluid-Structure Interaction (FSI) The most rigorous approach to modeling hydrodynamic forces solves the Navier-Stokes equations over a domain containing the vehicle and surrounding fluid. When