PaperlayerView exact revision on arXiv

Source-linked AI summary

Multifunctional Locomotion Control of Multi-Jointed BURs with Swimming and gait Capabilities

Takumi Asada, Hideo Furuhashi, Kenta Tabata, Renato Miyagusuku, Koichi Ozaki

arXiv:2609.37086v1cs.RO

TL;DR

The paper addresses locomotion control that depends on designer-selected behavior rules and limits new capabilities. It proposes a multi-jointed robot and sensor-responsive control, demonstrating transitions among as many as three behaviors. The authors expect this approach to expand underwater-exploration applications.

  • Problem

    Existing underwater-robot locomotion control relies on designer discretion and predefined behavior switching, while robots may be limited to target movements.

    Source: §1 · pp. 1, 3

  • Method

    The paper uses one multi-jointed structure with a two-layer CPG and potential-function control to generate sensor-responsive swimming and gait.

    Source: §1 · pp. 2–3, §3.1 · p. 7

  • Results

    Transitions among as many as three behaviors were demonstrated using estimated depth and visual information as sensor inputs.

    Source: §4.3 · p. 14, §5 · p. 16

  • Takeaways & Limitations

    The authors expect multi-jointed, multifunctional controllers to expand BUR applications in underwater exploration.

    Source: §5 · p. 16

    As the number of sensor modalities and behaviors increases, defining the motion-generation target becomes more complicated.

    Source: §4.3 · p. 14

Abstract

from arXiv · show

Multimodal biomimetic underwater robots (BURs) can conduct underwater tasks suitable for the environment. Combining the characteristics of aquatic organisms enables the swimming and leggedlocomotion required for underwater exploration. Locomotion control mechanism relies on rule-based behavior selection and the designer's discretion. This limits the robot's ability to acquire new behavioral capabilities to the predetermined range of behaviors. To address these challenges, we propose a mechanism and control system that enables the expression of multimodal locomotion capabilities from the same multi-jointed structure. A mechanism equipped with four leg-fins each having four axes is used. This controller achieves nonlinear behavior based on sensor modalities, rather than relying on predefined conditional rule-based on locomotion functions. This system was validated through both physical and simulation testing based on multiple sensor data and behavioral patterns. Utilizing a potential function in multimodal locomotion control was verified to enable transitions between two or three behaviors. Implementing the control method as a multimodal controller is expected to enhance its application in underwater exploration. Our project page is at https://tasada038.github.io/multi-jointed-bur/.

1. Introduction

Multifunctional underwater robots could broaden exploration, but existing approaches often rely on designer-selected structures and predefined behavior switching. This paper proposes generating swimming and gait from one multi-jointed structure, with locomotion responding nonlinearly to sensor modalities.

1. Introduction

  • Biomimetic underwater robots can be limited to their model organism’s movements, while thruster-propelled ROVs can disturb seafloor visibility through sediment upwelling.The paper frames multifunctionality as a way to address challenges faced by both types of underwater robot.Source: §1 · p. 1
  • The proposed control generates swimming and gait from the same multi-jointed structure, responding nonlinearly to sensor modalities rather than branching on locomotion function.The design aims to avoid separating or adding functions during the design phase.Source: §1 · pp. 2–3
  • Earlier approaches often predefined behaviors through designer-selected mechanisms and sensor-to-behavior mappings, limiting robots to target movements and complicating transitions.The cited approaches include threshold-based switching between swimming and gait.Source: §1 · p. 3
  • The manta ray robot was selected for its large pectoral fins, which serve as drive components for a multi-jointed, multifunctional design.The paper reports that manta rays move efficiently with minimal energy consumption due to their large pectoral fins.Source: §1 · p. 3

2. Structural Design of the Manta Ray Robot

The robot uses a four-leg-fin, multi-jointed structure with a two-layer CPG, plus variable buoyancy for depth adjustment. Its five-degree-of-freedom dynamic model includes estimated hydrodynamic terms and treats Coriolis and centripetal terms as negligible at slow speeds.

2.1. Hardware modeling

  • The robot’s four leg-fins each have four axes, and a two-layer CPG structure is used to generate different locomotion from the same multi-joint structure.The axes are yaw, roll1, pitch, and roll2; the CPG’s two layers are a single chain structure and a fully connected structure.Source: §2.1 · p. 4
  • A balloon variable buoyancy system adjusts depth; the body is heavier than neutral buoyancy by default, and gait requires torque for reliable surface push-off.The buoyancy system inflates and deflates to adjust depth and support underwater locomotion.Source: §2.1 · pp. 4–5

2.2. Dynamics of the manta ray robot

  • The dynamic model covers five degrees of freedom: surge, sway, heave, pitch, and yaw.The model uses earth-fixed and body-fixed coordinate systems to describe position, orientation, linear velocity, and angular velocity.Source: §2.2 · p. 5
  • At slow speeds, the model treats Coriolis and centripetal terms as negligible; its inertia matrix includes added mass and its damping matrix represents damping.The force and moment model uses n = 5.Source: §2.2 · p. 5
  • Added mass is estimated by summing contributions from the main body and leg-fins, using strip theory for the body and a flat-plate model for the fins.Damping coefficients are estimated from CFD results across inflow speeds of 0.4 to 1.2 m/s and pitching and yawing angles from -30° to 30°.Source: §2.2 · pp. 6–7

3. Multifunctional Locomotion Control

The controller combines a CPG-based motor execution system with a behavior transition system to generate swimming and gait from the same multi-jointed structure. Sensor modalities drive nonlinear behavior emergence rather than command-based on/off switching.

3.1. Motor execution system

  • CPG parameter changes generate swimming and gait from one multi-jointed structure, with two oscillator layers setting leg-fin motion and phase relationships.The single-chain layer connects each leg-fin to its lateral oscillators; the fully connected layer links each leg-fin base to every oscillator.Source: §3.1 · pp. 7–8

3.2. Behavior transition system

  • Depth estimation divides the water column into surface, mid, and deep categories, while RGB-D point-cloud clustering estimates surrounding terrain and wall presence.Depth uses trapezoidal and triangular membership functions; terrain perception uses DBSCAN with point-cloud normal-vector constraints.Source: §3.2 · p. 9
  • Depth determines the motion target: swimming is selected near the surface and gait in deep layers, while visual information weights different gait patterns.The controller estimates the robot state, selects a behavior-transition target, and executes motion based on that state.Source: §3.2 · p. 9
  • A Gaussian error potential in torus space guides transitions between target behaviors without multiple minima, preventing undesirable movement patterns during transitions.The potential gradient updates CPG parameters using current and target parameter values.Source: §3.2 · p. 10

4. Experiment and Result

Experiments demonstrate swimming and multiple gait motions from one multi-jointed structure, with potential-based control transitioning between behaviors. The system also shows how sensor modalities support three-behavior control, while defining motion targets becomes more complicated as modalities and behaviors increase.

4.1. Swimming ang gait locomotion

  • Swimming, trot, walk, and crab gait were generated from the same multi-jointed structure, with walk achieving the fastest speed and swimming speed trading off against gait speed.Measured position and velocity were approximately equal to simulation results.Source: §4.1 · p. 11

4.2. Two behavior transition of swimming and gait

  • Potential functions produced smooth swimming-to-trot transitions, with depth prediction determining the transition target when a single sensor modality is used.The reported swimming-to-gait transition occurred from 13 to 21 seconds.Source: §4.2 · pp. 12–13

4.3. Multifunctional generation of three behaviors

  • With predicted depth and visual information, the controller transitioned among swimming, trot gait, and turning gait; increasing modalities and behaviors complicates motion-target definition.Depth prediction selected swimming versus gait, while visual cluster results selected between trot and turning gait.Source: §4.3 · pp. 13–14

4.4. Potential applications

  • The controller uses sensor modalities and a simple multi-joint structure to generate multifunctionality without relying on specialized mechanisms or preconfigured functions.The paper presents this as extensible to other multi-joint robots, while noting that available sensor modalities are limited by the sensors a robot can mount.Source: §4.4 · pp. 14–15

5. CONCLUSIONS

The study reports multi-jointed BUR locomotion across four behaviors and demonstrates sensor-based transitions among up to three behaviors.

5. CONCLUSIONS

  • Swimming, trot gait, walk gait, and crab gait were generated by setting parameters in a two-layer CPG.Source: §5 · p. 15
  • The conclusion visuals cover locomotion snapshots and joint outputs, as well as wall-and-floor clustering results in water.Source: Figure 11 · p. 16, Figure 12 · p. 16
  • Transitions among up to three behaviors were demonstrated using estimated depth and visual information as sensor inputs.The controller uses a potential function for multifunctional behavior control.Source: §5 · p. 16

Funding

The authors report that no funding was received for the study.

Funding

Notes on contributor(s)

The contributor notes give the authors’ education, appointments, and research interests.

Notes on contributor(s)

  • Takumi Asada is a Ph.D. student at Utsunomiya University whose research interests include biomimetic underwater robots, nonlinear oscillators, and mechanical structure design.Source: Notes on contributor(s) · p. 17
  • Hideo Furuhashi works in electronics and electrical engineering at AIT and researches light, ultrasonic waves, robotics, and virtual reality.Source: Notes on contributor(s) · p. 17
  • Kenta Tabata is an assistant professor at Utsunomiya University; Renato Miyagusuku researches machine learning for robotics and sensor fusion, and Koichi Ozaki researches agricultural and mobile robotics.Source: Notes on contributor(s) · p. 17
Loading 2609.37086v1…