Hadal Small Vehicle (HSV)

• Hadal Small Vehicle (HSV) is an autonomous, compact submersible engineered for ultra-deep (6000+ m) operations with a pressure-tolerant hull and dual-medium capabilities.
• The design employs vectored thrust control, advanced PID navigation, and sensor fusion to achieve precise trajectory tracking with minimal errors.
• Integrated 3-DOF manipulators and suction cup end effectors demonstrate a 90% grasp success rate, ensuring reliable object recovery in harsh, high-pressure environments.

A Hadal Small Vehicle (HSV) is a compact autonomous underwater or aerial–submersible platform engineered for extreme operational environments in the hadal zone (depths beyond 6,000 m), where hydrostatic pressures can reach 60–110 MPa. Recent research has focused on both dedicated subsea AUV architectures and dual-medium vehicles capable of seamless air–water transitions. HSVs integrate pressure-tolerant hulls, vectored multirotor propulsion, sophisticated sensor suites, manipulator systems, and control frameworks that ensure reliable trajectory tracking and object recovery in structurally and physically hostile conditions (Grimaldi et al., 3 Jan 2026, Maia et al., 2015).

1. Mechanical and Hydrodynamic Design

HSV architectures feature a cylindrical pressure hull, dimensioned for survival at full ocean depth. For subsea variants, the pressure hull is paired with syntactic-foam buoyancy modules, optimizing neutral buoyancy by modeling both mass and volume compression under load. Hull volume reduction and center-of-buoyancy shifts are explicitly simulated as a function of depth. A nominal HSV measures ∼2.5 m in length, ∼0.6 m in diameter, and has a dry mass of ∼1,050 kg, accommodating payloads, notably three-degree-of-freedom manipulators and sensor/actuator electronics in the 100–150 kg range.

Thrust is delivered by eight vectored brushless DC thrusters arrayed symmetrically: four for horizontal surge/sway/yaw, and four for vertical heave/fine pitch control. Thruster allocation uses an $8\times4$ matrix $A$ to map body-frame wrenches ($[F_x, F_y, F_z, T_{yaw}]^T$) to individual thrust signals via $u = A \cdot F_b$, with saturation and rate limits enforced.

Hydrodynamics includes rigid-body plus added mass effects, Coriolis/centripetal coupling, linear and quadratic drag, and stochastic disturbance models. Drag computation employs experimentally derived coefficients:

$D(\nu) = D_1 \nu + D_2 |\nu| \nu$

where coefficients $D_1$, $D_2$ are vehicle-specific.

Dual-medium HSVs (Naviator-derived) extend this model with air/water force analysis, balancing lift, thrust, drag, and buoyancy envelopes such that:

$$L_{\max}(\omega_{\max,air}) \geq mg \ B(V_{tank}) = mg$$

for platform intersection and seamless operability in both media (Maia et al., 2015).

2. Manipulator and End Effector Systems

Hadal HSVs incorporate a three-degree-of-freedom electric manipulator. The architecture comprises:

• Base joint ($\theta_1$) for horizontal swing
• Shoulder joint ($\theta_2$) for elevation
• Elbow joint ($\theta_3$) for flexion/extension

Link lengths are typically $l_1 \approx 1.0\,\mathrm{m}$, $l_2 \approx 0.56\,\mathrm{m}$, with joint torque ratings at $\sim200\,\mathrm{N\cdot m}$ peak for full-depth operation. Max reach is $\sim1.5$ m horizontally and $\pm1.2$ m vertically.

The end effector is a suction cup with integrated micro-pump, abstracted in simulation (Stonefish “glue” constraint) such that successful attachment occurs within $1\,$cm alignment tolerance. This coupling omits direct fluid–structure and pump flow dynamics; future work will incorporate CFD-based seal modeling.

Manipulator–vehicle coupling is fully modeled: arm inertia, center-of-mass shifts, and reaction-torques propagate to—the hull, preserving dynamic fidelity.

3. Simulation Environment and Control Frameworks

Realistic system evaluation is performed in the Stonefish simulation environment. Full six-DOF rigid-body dynamics, depth-adjusted mass/inertia, hydrodynamic disturbance (currents, eddies), and sensor degradation as a function of depth are integrated. The principal dynamic equation:

$$M_{RB}\,\dot{\nu} + C_{RB}(\nu)\,\nu + M_A\,\dot{\nu} + C_A(\nu)\,\nu + D(\nu)\,\nu + g(\eta) = \tau + \tau_{dist}$$

For multirotor dual-medium designs, Newton–Euler equations govern both air and water regimes, using experimental thrust curves for propeller and motor efficiency:

$T_{i} = C_{T}\,\rho\,A_{disk}\,\omega_{i}^{2}$

where $\rho$ is the medium density.

Vehicle navigation uses world-frame PID controllers. Position and yaw errors are

$e_p = p_{wp} - p = [e_x,\,e_y,\,e_z,\,e_\psi]^T$

with per-axis control laws,

$$U_i(t) = K_{p,i}e_i(t) + K_{i,i}\int_{0}^{t} e_i(\tau)d\tau + K_{d,i} \frac{de_i}{dt}$$

Manipulator control employs geometric inverse kinematics and acceleration-level PD feed-forward:

$$\ddot{q}_{cmd} = \ddot{q}_d + K_v (\dot{q}_d - \dot{q}) + K_p (q_d - q)$$

4. Sensor Suite and Perception Methods

HSV sensor packages for deep-sea recovery consist of:

• 10 Hz IMU (accel/gyro), Doppler Velocity Log (DVL) for ground-relative velocities
• Depth (pressure) sensor with ±0.1 m RMSE
• Fused INS using EKF for state estimation
• Down-looking monochrome camera array augmented with LED for illuminated visual detection
• Forward-looking stereo cameras (installed, not used in some studies)

Vision modules segment targets (e.g., starfish-shaped object) from the camera stream; centroid/orientation are utilized by manipulation planners. Sensor noise/bias increase with depth are systematically incorporated in simulation.

5. Mission Profiles and Performance Metrics

The standard hadal recovery sequence commences with autonomous descent to 6,000 m, waypoint-driven seafloor coverage, real-time visual search using the downward camera, stabilization upon target detection, arm deployment to a pre-grasp pose $~5$ cm above object, suction cup attachment, object lift ($\sim0.3$ m), retraction, and either resurfacing or continuation.

Reported simulation metrics:

Metric	Value/Comment	Trials
RMS horiz. position error	~0.4 m (lawnmower survey at 6,000 m)	All runs
Depth error	±0.1 m of setpoint	All runs
Yaw error	<2° (coverage/station-keeping)	All runs
INS drift (30 min)	~1.2 m cumul., N: 0.8 m, E: 1.0 m	Survey
Manipulator alignment	≤2 cm in 9/10 runs (failure: flow swirl)	10 runs
Suction attachment success	9/10 (failure: seal misalignment)	10 runs
Currents (random, up to 0.1 m/s)	Induced base motion peaks of 0.3 m pre-PID correction	All runs
Yaw deviation (arm inertial)	Up to 1.5° (controller maintained stability)	All runs

Autonomous recovery demonstrated high reliability (90% grasp success), precise station-keeping, and effective mitigation of hydrodynamic disturbances.

6. Design Optimization and Adaptation for Hadal Operations

Design trade-offs are centered on lift/thrust versus weight, buoyancy versus drag, and hull strength versus mass. Structural survival dictates material and thickness selection, e.g., Ti 6Al‐4V pressure hulls with $t\geq 19.5\,$mm for $r=0.08\,$m under $p=110$ MPa. Syntactic foam and internal bulkheads minimize flexural failures; electronics are epoxy-potted to 200 MPa.

Optimization targets minimize total mass for required buoyancy and flight performance: $$\min_{m_s,m_b,V_t}\; m_s + m_b\quad \text{s.t.}\quad V_t\,\rho_w = m_{\text{total}},\; \sum_i C_{T,a}\,\rho_a A\,\omega_i^2 \geq m_{\text{total}}g$$ A prototypical solution is $m_{struct}\sim1.0\,\mathrm{kg}$, $m_{batt}\sim0.6\,\mathrm{kg}$, $V_{tank}\sim1.6\,\mathrm{L}$, $m_{total}\sim2.2\,\mathrm{kg}$, yielding a hover thrust margin $\sim20\%$ (Maia et al., 2015).

Dual-medium HSVs demand rapid air–water transition protocols: blades crossing the interface shut down to prevent drag spikes, compensated by remaining rotors. Stability requires thrust asymmetry $|\Delta T|\leq T_{hover}/2$.

7. Limitations, Failure Modes, and Future Work

Simulation-based validation may overlook complex fluid–structure and pump dynamics, soil embedment mechanics, and non-ideal visual conditions (turbidity, illumination attenuation). IMU drift without aiding (e.g., USBL or bathymetric SLAM) can cause waypoint loss over extended missions. Component failures (thruster/joint) reduce control authority; manipulator misalignment under local vortices can cause suction failure.

Field deployment is expected to benefit from compliant, multi-DOF end effectors and absolute aiding for large-scale surveys. Future work will focus on CFD-based suction modeling, soft-manipulator component integration, expanded autonomy stacks, digital-twin health monitoring, multi-agent subsea coordination, and progressive hardware-in-the-loop/shallow-water test validation (Grimaldi et al., 3 Jan 2026).

A plausible implication is that integration of high-fidelity simulation and robust control law tuning is essential before full-scale ocean trials, given the substantial risk and operational costs associated with hadal-depth experimentation. Further advancements will address sensor reliability, environmental disturbance rejection, and precision manipulation for comprehensive hadal-zone science and object recovery.