Underactuated Biomimetic Underwater Robot

Updated 16 November 2025

Underactuated biomimetic underwater robots are systems that emulate biological swimmers by using fewer actuators than degrees of freedom, leveraging compliant and passive mechanisms.
They incorporate innovative mechanical designs like bistable elements, compliant joints, and passive linkages to achieve efficient propulsion, maneuverability, and manipulation.
Their control strategies span from open-loop mechanical methods to reinforcement learning and hybrid adaptive feedback, enabling energy-efficient performance in complex aquatic tasks.

An underactuated biomimetic underwater robot is a class of autonomous or remotely operated vehicles designed to emulate biological swimmers using fewer actuators than kinematic degrees of freedom (DOF), often relying on compliant, passive, or mechanically intelligent structures to enable efficient propulsion, maneuverability, or manipulation. These robots leverage biological principles such as underactuated fin or tail mechanics, bistable elements, and compliant morphologies to achieve complex, high-performance motion or manipulation in underwater environments while minimizing actuation and control complexity.

1. Mechanical Architectures and Underactuated Schemes

Underactuated biomimetic underwater robots span a spectrum from fish-like swimmers (e.g., "CarbonFish" (Xiong et al., 2023), fish-mimetic AUVs (Singh et al., 9 Nov 2025), and fin-driven AUVs (Remmas et al., 26 Apr 2025)) to multi-limbed cephalopod analogs (octopus-like robots (Zhang et al., 15 Oct 2024)) and compliant graspers inspired by biological mouths (Yoshimura-folding grippers (Guo et al., 14 Mar 2025)). Common to all is the decoupling of actuation input from total system DOF through one or more of:

Minimal direct actuation: e.g., a single actuator in a compliant tail (four-joint, n=4 tail, m=1 actuation at distal joint (Singh et al., 9 Nov 2025)), two motors driving eight arms in parallel via passive linkages (octopus robot (Zhang et al., 15 Oct 2024)), or one degree-of-freedom cable/gripper closing (origami gripper (Guo et al., 14 Mar 2025)).
Bistable and compliant elements: The bistable Hair-Clip Mechanism (HCM) in CarbonFish leverages CFRP ribbons prestressed across a defined width, creating snap-through energy barriers between stable configurations and enabling efficient energy transfer on actuation (see $\ell$ , $h$ , $t$ , $D$ geometry and $E$ material modulus, Table 1).
Passive compliance and mechanical intelligence: Elastomeric joints, variable stiffness segments (incisions in soft arms (Zhang et al., 15 Oct 2024)), and compliant creased structures (Yoshimura origami) are used to ensure that body or appendage segments adapt to hydrodynamic loads, serve as antagonistic energy storage, and reduce energetic demands on actuators.

Table 1: Representative Mechanical Configurations

Robot/Paper	Actuator Count / DOF	Main Materials/Mechanism
CarbonFish (Xiong et al., 2023)	1 actuator, 2 DOF (tail)	CFRP hair-clip ribbons, bistable
Fish-like AUV (Singh et al., 9 Nov 2025)	1 actuator, 4 DOF (tail)	3D-printed & Al body, elastomer joints
Octopus robot (Zhang et al., 15 Oct 2024)	2 motors, 8 arms	Silicone arms, offset crank-slider linkage
Origami gripper (Guo et al., 14 Mar 2025)	1 cable, 1 DOF (grip)	PET/PETG laminate, Yoshimura crease fold
Fin-driven AUV (Remmas et al., 26 Apr 2025)	4 motors, 6 DOF vehicle	Rigid oscillating fins, symmetric mounting

2. Dynamic Modeling, Kinematics, and Energy Transfer

The dynamic and kinematic models of underactuated biomimetic robots universally blend reduced-order actuation with compliance and environmental coupling.

Bistable elements: In CarbonFish, the out-of-plane ribbon deflection $\varphi(z)$ follows a Bessel-function-based Euler beam model, with snap-through energy $U_\mathrm{barr} \approx 3 P_\mathrm{cr} D$ defining actuation events. The inertial-elastic time scale and actuator servo limit the maximum undulation frequency (up to 10 Hz experimentally).
Multibody tail models: For fish-like AUVs, the equations of motion are $M(q)\ddot{q} + C(q,\dot{q})\dot{q} + Kq + F_\mathrm{hydro}(\dot{q}) = Bu$ , with only the distal joint actuated. Upstream joints respond passively according to their elastic and hydrodynamic characteristics.
Complex manipulators: Origami grippers translate a single imposed linear displacement $u$ into a nonlinear aperture variation, $\theta(u)$ , via the Yoshimura fold, with grasp forces determined by panel geometry, crease pattern, and pulling force analytic relations.
Fin and foil arrays: Fin-driven AUVs are modeled using thruster allocation via instantaneous and cycle-averaged thrust/torque mapping, with the force–moment mapping conducted through adjoint and block-diagonal transformations of fin torques into body-frame wrenches.

3. Control Methodologies: Open-Loop, RL, and Hybrid Adaptive Schemes

Control frameworks for underactuated biomimetic robots are tailored to reconcile minimal actuation with required behavioral richness.

Open-loop mechanical intelligence: The octopus robot achieves swimming and planar steering via constant-speed motors and an asymmetric quick-return crank-slider (stroke ratio $K$ tuneable from 1.2:1 to 2:1), with differential motor speeds generating yaw without explicit high-level feedback.
Reinforcement learning (RL) for efficiency: In (Singh et al., 9 Nov 2025), the underactuated fish AUV is controlled via Proximal Policy Optimization (PPO), where the policy network outputs torque commands at the actuated tail joint. The reward function balances forward velocity and energy use, converging to minimal-energy, robust behaviors in simulation.
Hybrid adaptive feedback: For multi-fin AUVs, the hybrid controller integrates quaternion-based pose error metrics, Lyapunov-stable switching, and online parameter adaptation; the resultant desired wrench is passed through analytic control allocation (CA_prop), which inversely maps wrench demands to fin angles and amplitudes (sub-100 µs allocation time on Jetson TX2 (Remmas et al., 26 Apr 2025)).

4. Experimental Performance and Comparative Metrics

Quantitative evaluation demonstrates the capabilities and limitations of underactuated biomimetic robots relative to their biological and robotic counterparts.

Speed and frequency: CarbonFish achieves undulation frequencies up to 10 Hz and swimming speeds up to $\sim 8$ body lengths per second (BL/s), with a projected maximum of 6.8–10.8 BL/s—substantially higher than soft robots or real fish baseline (0.3–0.7 BL/s; 2–4 Hz) (Xiong et al., 2023).
Efficiency and thrust: CarbonFish exhibits a hydrodynamic thrust of $\approx 0.2$ N at 10 Hz and mechanical efficiency of $\approx 12$ \%. The underactuated fish AUV achieves 70% of the fully actuated model's cruising speed (0.18 vs. 0.20 m/s) while using only 65% of the energy per unit distance (5.5 vs. 8.0 J/m) (Singh et al., 9 Nov 2025), indicating superior energy efficiency with minimal active control.
Control accuracy: The four-fin AUV attains centimeter-scale 3D tracking accuracy (RMSE 0.04 m linear, 0.06 rad angular) in simulation and $\sim$ 8.5 cm depth RMSE in pool trials, substantially outperforming classical allocation benchmarks (Remmas et al., 26 Apr 2025).
Manipulation robustness: The fish-mouth origami gripper can exert envelope grasp forces up to $\sim$ 4.5 N and pinch forces of $\sim$ 3 N at up to 0.5 s closure times, with compliance and robustness validated across >300 underwater grasping trials (Guo et al., 14 Mar 2025).
Octopus robot swimming: The umbrella-inspired octopus robot attains peak upward velocities of 314 mm/s, exceeding alternative designs by 3.5–8x, with propulsive efficiency on the order of 10–20% (Zhang et al., 15 Oct 2024).

5. Applications: Exploration, Monitoring, and Manipulation

Underactuated biomimetic underwater robots are utilized in diverse aquatic missions:

Ecosystem monitoring: The fish-like AUV incorporates minimal actuation and compliance to reduce acoustic/hydrodynamic signatures for stealthy habitat transects, coral health assessments, and water-quality profiling in sensitive aquatic environments (Singh et al., 9 Nov 2025).
Manipulation and grasping: Origami grippers have been deployed for gentle, robust grasping of varied marine specimens and objects, including delicate invertebrates and simultaneous grasping of multiple items (Guo et al., 14 Mar 2025).
Trajectory tracking and inspection: Fin-driven AUVs employ adaptive and computationally efficient controllers for complex 3D tracking tasks, with resource usage suitable for real-world embedded deployment in mapping or surveillance missions (Remmas et al., 26 Apr 2025).
Biological model replication: Octopus-mimetic platforms enable interdisciplinary paper of cephalopod biomechanics and are promising for further development of highly maneuverable, compliant underwater explorers (Zhang et al., 15 Oct 2024).

6. Design Limitations and Perspectives for Future Development

Open technical challenges and future directions encompass:

Actuation bandwidth: Servo/motor speed is frequently the limiting factor for undulation frequency (e.g., CarbonFish), suggesting the integration of higher-speed motors or motor–gearbox assemblies for improved performance (Xiong et al., 2023).
Compliant sealing and structural durability: Open-frame or multi-material assemblies pose waterproofing and fatigue life challenges; advanced overmolding, flexible boots, or composite material selection are directions for robust system longevity.
Enhanced continuum kinematics: Current underactuated designs often approximate multi-segment waveforms with a small number of discrete curvature nodes or segments. Increasing the segment number and incorporating variable phase lag or adaptive stiffness elements would more closely mimic biological swimming and potentially enhance efficiency or agility.
Learning and adaptive control transfer: RL policy transfer from simulation (e.g., FishGym) to physical vehicles and expansion to multi-joint underactuated architectures remain active areas for practical deployment (Singh et al., 9 Nov 2025).
Integrated sensing and autonomy: Incorporation of low-power, miniaturized in situ sensors (e.g., CTD, pH, stereo vision) and robust autonomy platforms will be crucial for fully realizing the ecological survey and intervention potential of these platforms.

7. Significance and Interface with Broader Research

The evolution of underactuated biomimetic underwater robots demonstrates the effectiveness of exploiting compliance, morphological computation, and minimal-actuator design for achieving high-performance aquatic locomotion and manipulation. These approaches leverage a combination of advanced materials (CFRP, silicone, PETG laminates), analytic modeling, and emerging reinforcement and hybrid control frameworks to approach or surpass the capabilities of legacy underwater vehicles, often with substantial gains in efficiency, accuracy, or mechanical simplicity. Their continued development offers the potential for robust, adaptable, and energy-efficient autonomous systems for exploration, monitoring, and manipulation across a variety of underwater domains.