IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO.

5, OCTOBER 2012

847

On a Bio-inspired Amphibious Robot Capable of Multimodal Motion

Junzhi Yu, Rui Ding, Qinghai Yang, Min Tan, Weibing Wang, and Jianwei Zhang
AbstractThis paper addresses the system design and locomotion control for a versatile amphibious robot, AmphiRobot-II, inspired by various amphibian principles in the animal kingdom. In terms of the propulsion features of existing amphibians, a novel hybrid propulsive mechanism coupled with wheel-propeller-n movements is proposed that integrates sh- or dolphin-like swimming and wheel-based crawling. The robot is able not only to implement exible wheel-based movements on land, but also to perform steady and efcient sh- or dolphin-like swimming under water and can further switch between these two patterns via a specialized swivel device. To achieve multimodal motions, a body deformation steering approach is proposed for the turning locomotion on land with minimum turning radius obtained accordingly. A central pattern generator inspired underwater locomotion control is also implemented and tested on the physical robot. Based on the aforementioned design, the AmphiRobot-II prototype has been built and has successfully demonstrated to conrm the effectiveness of the hybrid propulsive scheme and the amphibious control approaches. Index TermsAmphibious robot, bio-inspired robot, mechanical design, motion control, multimodal motion.

I. INTRODUCTION

IO-INSPIRED mechatronics is emerging as a multidisciplinary subject focusing on learning and mimicking biologic characteristics and functions and then reproducing and even outperforming these characteristics [1] [3]. It has drawn great interest in the area of robotics related to biological functions, mechatronics, control technology, computational intelligence, etc. One of the crucial features that an amphibious robot should exhibit is multiple mobility suitable for diverse working conditions. A primary motivation is that most existing robots

Manuscript received July 12, 2010; revised November 22, 2010; accepted February 12, 2011. Date of publication May 10, 2011; date of current version August 17, 2012. Recommended by Technical Editor G. Morel. This work was supported in part by the National Science Foundation of China under Grant 60725309 and Grant 61075102, in part by the Beijing Natural Science Foundation under Grant 4082031 and Grant 4102063, in part by the Nation 863 Program under Grant 2007AA04Z202, and in part by the Alexander von Humboldt Foundation of Germany. J. Yu, R. Ding, and M. Tan are with the State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China (e-mail: [email protected]; [email protected]; [email protected]). Q. Yang is with the Research Institute of Petroleum Exploration and Development, Petrochina, Beijing 100083, China (e-mail: [email protected]). W. Wang is with the Machine and Electricity Engineering College, Shihezi University, Shihezi 832003, China (e-mail: [email protected]). J. Zhang is with the Department of Informatics, University of Hamburg, 22527 Hamburg, Germany (e-mail: [email protected]). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TMECH.2011.2132732

can only operate in a specic environment. For example, the ground-based robots are unable to work in water due to lack of a working aquatic propulsor and necessary waterproof device, while the predominant underwater robots lack sufcient terrestrial motion ability as they will undergo larger friction on land. As a consequence, inspired by amphibians such as the crab, tortoise, frog, crocodile, penguin, and newt in nature, researchers show an intense desire to design and develop autonomous, selfcontained amphibious robots not only suitable for the land as well as turbulent ocean surf zones, but also adaptable to complicated aquatic circumstances [4][6]. The development of fully autonomous amphibious robots has great theoretical value in exploring biological design principles as well as broad application prospects related to underwater environment exploration and data gathering with minimum disturbance to the surrounding life-forms. The existing amphibious robots, in general, fall into two categories in terms of the applied locomotor propulsor, i.e., leg-like amphibious robots and snake-like ones. The former representatively involves the cockroach-inspired wheel-leg propellers named Whegs [7], a lobster-like robot for the purpose of neural control [8], the crab-like autonomous legged underwater vehicles [9], etc. Attempts have also been made to simulate anguilliform swimming by sea snakes and lampreys in water and lateral undulatory locomotion by snakes on land. Such snake robots with special attention to waterproof protection include the modularly designed ACM-R5 by the Tokyo Institute of Technology [10], and the AmphiBot developed by the Swiss Federal Institute of Technology [11]. As two distinct locomotor forms, the leg- and snake-like propulsors have their own characteristics suitable for varying applications. Conventional legged robots should touch the land and capitalize on mutual friction for forward motion, which is primarily applicable to movements on the land or the sea bottom. By contrast, common snake robots only perform planar crawling motions, where 3-D swimming can be accomplished only when specic modications are made to the propulsive structure. Furthermore, unlike legged robots that may even perform well in rugged terrain, snake robots might get partly or completely stuck in soft wetland that has an adverse impact on the ongoing locomotion. On the other hand, leg-like propulsors usually pose a great challenge to maintenance overhaul owing to the complicated mechanical conguration. Then again, slender snake robots are easily manufactured as a modular structure. So there is a serious need to develop a versatile amphibious robot affording a plurality of primary locomotion means. The aim of this paper, on the basis of our previous research on the amphibious robot [12], [13], is to provide an updated version for more versatile and excellent locomotor performance

1083-4435/$26.00 2011 IEEE

848

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO. 5, OCTOBER 2012

both on land and in water. The robots key features reside in: 1) modularly designed sh-like propelling units according to the philosophy of modularization and morphological mimicry are utilized to achieve fast and exible swimming; 2) the incorporation of a pair of composite wheel-propeller-n mechanisms into the robot AmphiRobot-II ensures a good performance of both terrestrial and underwater locomotion; 3) a specially designed swivel mechanism allows a smooth transition between the sh- and dolphin-like modes; and 4) a body-deformationbased steering approach is proposed for the terrestrial motions, whereas a central pattern generator (CPG)-based locomotion control method is used for swimming motions. Furthermore, all mechanical designs and control methods are veried by extensive on-site experiments. According to the authors knowledge, the AmphiRobot-II may be the only miniature amphibious prototype capable of sh-like swimming, dolphin-like swimming, and a propeller mode for propulsion under water. The rest of the paper is organized as follows. An overall description of the updated mechanical design scheme for the AmphiRobot-II together with its implementation is provided in Section II. The body-deformation-based steering approach is proposed for terrestrial locomotion in Section III. The CPG-centered underwater locomotion control is detailed in Section IV. Experiments and analysis performed are presented in Section V. Finally, conclusions and future work are summarized in Section VI. II. UPDATED MECHANICAL DESIGN AND REALIZATION A. Updated Scheme Our previous work on amphibious locomotion focused on proof-of-concept propulsive mechanisms and resulted in a robot AmphiRobot-I [12], which partly possesses scheduled terrestrial and/or aquatic motions. However, there exist many open problems that degrade the propulsive performance to some extent. 1) Compared to the biological counterpart, the oversimplication of the AmphiRobot-I regarding shape generates adverse hydrodynamic drag when swimming in water. 2) The clearance between the adjacent sh-like propelling units that will be lled with uids is somewhat big, which will slow the forward speed. 3) Due to the insufcient torque produced by the adopted dc motor, the devised wheel paddle mounted on the forehead is unable to drive the robot quickly on land. 4) The overemphasized modular concept leads to the underestimation of the gross mass of the last propelling unit, including the caudal peduncle and toothed synchro belt, accompanied by insufcient buoyancy. Taking account of these open problems, we proposed an updated concept design of the AmphiRobot-II as shown in Fig. 1, which is capable of effective multimodal gaits both on land and in water. The oscillations of a chain of sh-like propelling units in conjunction with a composite peduncle mechanism and a caudal n perform carangiform swimming. The number of modular propelling units can be appended or detached depending on the demand. Notice that only the last third of the body length of a

Fig. 1. Mechanism design of AmphiRobot-II. (a) Concept design. (b) Prototype of the amphibious robot in eld test.

carangiform swimmer participates in undulations in the context of sh swimming [14]. As the most unique point of this robot, it has a joint with roll-axis at the neck (i.e., a swivel mechanism), which enables the robot to change the axis of posterior joints between yaw axis and pitch axis. Thus, using this joint, the robot can perform both sh-like and dolphin-like swimming. Additionally, with the multipurpose ipper as an antetype, a pair of composite wheel-propeller-n mechanisms driven by hightorque motors is introduced and fabricated. It acts as the wheel for crawling on land and as the pectoral n for balance and propulsion in water. Moreover, some engineering issues related to buoyancy design, component layout, and waterproof sealing are further coped with effectively. B. Morphological Modication During swimming the complex hydrodynamic interaction still remains an open problem, where a well-streamlined body plays an important part in vortices-based control and drag reduction. In terms of hydrodynamics, form drag caused by the distortion of ow around solid bodies is determined by their shapes. Therefore, we have effected some modications. 1) To reduce or eliminate added mass effect, the clearance between neighboring propelling units is lled with lightweight convexo-concave cork that can reduce the fore-and-aft space, as shown in Fig. 2(a). Incidentally, an interspace of 3 mm is held at the current version. 2) As illustrated in Fig. 2(b), the posterior clearance between the last module and the peduncle is crammed with a foam structure with an inll of aluminium sheathing, which can ll up the deciency of buoyancy and generate added propulsion due to interaction with expelled uid. 3) Instead of the at prole of the AmphiRobot-I, the side panels of the sh-like propelling unit made of transparent perspex are changed to convex camber so as to offer

YU et al.: ON A BIO-INSPIRED AMPHIBIOUS ROBOT CAPABLE OF MULTIMODAL MOTION

849

Fig. 2. Morphological modication in AmphiRobot-II. (a) Reduced fore-andaft clearance between sh-like modules. (b) Reduced clearance between the third module and the peduncle (shadow area).

less resistance to uid ow as well as a more picturesque prole [see Fig. 1(b)]. C. Wheel-Propeller-Fin Mechanism Driving multiple separate propulsive elements via a shaft within a tight space is a technical challenge. For the preceding AmphiRobot-I, the ipper and wheel that are mounted on the output shaft of the dc motor (Maxon EC-max 16/GS16A) are interchangeable. The reciprocating apping ipper acts as a pectoral n assisting in propulsion and maneuverability in water, and as a limb for slow crawling on land. The variant wheel with four spokes and four feet does not drive the robot fast enough into the rolling mode as it lacks sufcient motor torque. No doubt, it is very inconvenient to manually switch the motion modes between the apping ipper and the rolling wheel, especially inappropriate for autonomous locomotion in ever-changing environmental conditions. Moreover, the ipper actuated by the dc motor is not very suitable for fast response apping due to low mechanical efciency caused by frequent startingacceleratingbrakingreversing in actuators. Additionally, to accomplish an overall cycle of online trajectory control with an allowance of consuming time for planning and decision, longer time is required, which deteriorates the maximum locomotion potential of the ippers. Taking all these factors into consideration, we propose a novel wheel-propeller-n mechanism where a unique coaxial shaft is utilized to drive the wheel propeller and the ipper separately. The employment of this hybrid mechanism permits a distributed control of terrestrial and underwater locomotion simultaneously that guarantees a better performance and more autonomous switch during locomotion. As illustrated in Fig. 3, the composite coaxial shaft consists of two independent outputs: the inner shaft and the outer shaft. The former actuated by a servo drives the articial ipper, while the latter actuated by the dc motor drives the wheel propeller. In particular, an extension xture that has been utilized as a symmetrically assembled coaxial shaft will laterally take up too much room in the head unit (only 150 mm in width). An axlebox with a custom-made sealing unit is utilized to combine the

Fig. 3. Integrated wheel-propeller-n mechanism. (a) Back view. (b) Front view. (c) Mechanical conguration of a composite coaxial shaft. (d) Manufactured parts.

composite shaft with the outside of the side panel by mechanical means. The actuators, i.e., dc motor and servo, are vertically arranged on the inside of the side panel rather than on the bottom of the head adopted in the previous version, which highlights the modular concept and simplies the assembly [see Fig. 3(a)]. As a bonus offered by the new version, a gear set with a reduction ratio of 1:2 is utilized by the servo for driving the inner shaft, which expands the movable range from 180 to 360 (noncontinuous) and especially enables the ipper to reversely ap for backward swimming. Moreover, up-and-down locomotion can also be achieved by imposing an angle of attack on the pectoral n. As for the dc motor (Maxon RE-Max 24 + GP22C with a reduction ratio of 53:1) with larger torque, it drives the outer shaft through a bevel gear set corresponding to a revolution direction change of 90 . The connected wheel propellers acting as drive wheels, in conjunction with a pair of passive wheels attached to the base of the last sh-like propelling unit, achieve fast-moving roll, forward or backward. When spinning underwater, the specialized four spokes [see Fig. 3(b)] function as a common screw propeller that can steer the robot by lateral thrust vertical to the fore-and-aft body, i.e., the wheel propeller can be independently used for lateral maneuvers. As a side effect brought about by the modular design, waterproof-related issues of the wheel-propeller n should be fully considered. To this end, we have devised a specialized sealing unit based on mechanically dynamic seals. The cavum between the outer shaft and the axlebox, whose ends are xed by bearings, is lled with nonhydrophilic, frictionless grease. The inner shaft is positioned and sealed in like manner. Ultimately, the side panel equipped with the wheel-propeller-n mechanism is mounted into the head via screwing as a module, which facilitates repeated disassembly and assembly. Therefore, the wheel-propeller-n mechanism can be regarded as a multipurpose propelling tool.

850

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO. 5, OCTOBER 2012

TABLE I TECHNICAL SPECIFICATION OF THE DEVELOPED AMPHIROBOT-II

Fig. 4. Newly designed clutch for the swivel mechanism. (a) Mechanical sketch. (b) Manufactured parts.

D. Swivel Mechanism Beneting from the specially designed swivel mechanism, the robot can realize distinctive swimming modes combining shand dolphin-like swimming. In the previous version, the servo output shaft behind the head is inserted into the hole in the front end of the rst modular unit and xed by a countersunk rivet screwing vertically from above. Due to the mismachining tolerance as well as the existing t clearance, a nonsmooth switch during swimming, especially an imprecise position transition (not 90 in the strict sense), will induce the inclination of the robots body. This inclination badly affects the locomotion performance. Hence, a newly designed clutch structure shown in Fig. 4 has been exploited to overcome the potential contrariousness. The clutch is welded to the output shaft and two additional small axes distributed on contralateral sides can assist the output shaft in positioning. The prominent improvement lies in the line connection in place of the point connection. This permits stronger xing, which facilitates a more smooth switch between sh- and dolphin-like modes. E. Mechanical Implementation Based on the aforementioned mechanism design scheme, an updated amphibious robot has been developed. As depicted in Fig. 1(b), the robotic prototype with unique wheel-propeller ns mainly comprises a head unit, three sh-like propelling units, and a composite peduncle mechanism followed by a crescent tail n. To imitate n elasticity exhibited in sh, a soft elastoplastic material, Roylar, is used for fabricating the tail n. Meanwhile, more rigid Nylon 66 is exploited for the pectoral ns. The corresponding technical parameters of the AmphiRobot-II are listed in Table I. III. TERRESTRIAL LOCOMOTION CONTROL A. Body-Deformation-Based Steering Characteristics As a primary motion pattern, terrestrial locomotion control involves good stability and maneuverability as well as robust obstacle avoidance [15][17]. According to the mechanical structure of the AmphiRobot equipped with wheel paddles (AmphiRobot-I) or wheel-propeller ns (AmphiRobot-II), wheeled locomotion is the basic mode on land. Taking the body structure of the AmphiRobot into careful consideration, the drive of the AmphiRobot is more similar to that of a car drive. The perpendiculars of two mutually independent wheels and the xed wheels of a car form an instantaneous center of rotation (ICR) that induces the orientation adjustment of the car. When the robot body remains straight, the fore driving wheel-like part and rear passive wheels are parallel, forming no ICR, and the robot moves forward. Beneting from the carangiform swimming in water, the robots body shape can be varied when the modular propelling units depart from their central positions. Then the perpendiculars of wheel paddles and passive wheels intersect where an ICR is formed and the robot can maneuver on land. Such a maneuvering procedure is, hereinafter, referred to as body-deformation steering. For our robot with three sh-like propelling units, the independent rotation of the second or third unit or their coordinated deections will yield a deformed body shape that will form an ICR. Therefore, three types of deformation can be deployed to steer the robot agilely on land. These three methods form different ICRs corresponding to different turning radii, which will be discussed later. B. Geometry-Based Deformation for Steering As shown in Fig. 5(a), both the second unit and the third one depart from their middle positions with offsets of and , respectively. The body shape turns from a straight line into an approximate arc corresponding to an ICR. Once, the deection angles are given, D1 and D2 can be calculated based on the geometrical relationship of the triangle P QX D D1 = sin sin ( ) D2 D = (1) sin sin ( )

where D is a constant. As the two right triangles share the same hypotenuse, one can obtain sin( + ) = sin = L 2 L
L1 L

L2 L1 = sin( + ) sin

(2)

where L1 = D1 + D3 , L2 = D2 + D4 , while D3 and D4 are determined by the prototype. Combining (1) and (2), can be solved. Then the corresponding turning radius can be expressed

YU et al.: ON A BIO-INSPIRED AMPHIBIOUS ROBOT CAPABLE OF MULTIMODAL MOTION

851

Fig. 6.

Comparative result for three methods to form ICR.

Fig. 5. Illustration of changing the body shape via deections of the propelling units to form an ICR. (a) Coordinated deections of the second and third propelling units. (b) Separate rotation of the second unit. (c) Separate rotation of the third unit.

by R = L1 cot( + ) R= D (sin + sin cos ( + )) + D4 sin ( + ) sin2 ( + ) + D3 sin ( + ) cos ( + ) . sin2 ( + ) (3)

As a special case shown in Fig. 5(b) and (c), the independent rotation of the second or third unit will also form an ICR. In like wise, the following relations result: D + D3 cos + D4 R = sin (D + D3 ) cos + D4 R = . sin (4) (5)

Fig. 7.

Body deformation of a modular robot with N similar modular units.

In order to obtain the optimal turning mode, the radii of three turning approaches with the same deection angles will be analytically compared. Assume that + = = (6)

where the superscripts correspond to the two particular cases. It geometrically follows that the inequality holds as R < R < R . Fig. 6 plots the comparative result. As can be seen, with the same deection, the case employing the third propelling unit to rotate around its spin axis will generate the minimal turning radius. Additionally, the individual rotation of the third unit causes less deviation of passive wheels from the longitudinal centerline, which enhances the robots stability. C. Extension of the Body-Deformation Steering In light of the modular structure of the AmphiRobot, we can expand the conclusion to a modular amphibious robot with multiple joints as a supplement. Assume that the robot has N

modular units as shown in Fig. 7, and the deection angles are denoted by 1 , 2 , . . . , N 1 , where i corresponds to the joint angle of the (i + 1)th modular unit relative to the ith one. The deformation of N modular units can be viewed as an N -polygon with N sequentially connected links, and the rst modular unit is still attached to the head that constitutes the rst link in union. The lengths of links are denoted by L1 , L2 , . . . , LN , with (xi1 , yi1 ) and (xi , yi ) as the start and end points of the ith link. Here L1 denotes the length between (x0 , y0 ) and the output shaft of the rst unit (corresponds to (x1 , y1 )). The coordinate system is established in Fig. 7, where (x0 , y0 ) corresponds to the origin (0, 0). As the midpoint of the two rear passive wheels corresponds to the end point (xN , yN ), the ICR is the intersection point of the X-axis and the perpendicular line orthogonal to the ith link while passing (xN , yN ). The turning radius is then the distance between the ICR and the origin of the coordinate. From Fig. 7, the end point (xN , yN ) can be expressed as follows: xN = N Li sin( i1 j ) i=2 j =1 yN = L1 + N Li cos( i1 j ). i=2 j =1 (7)

852

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO. 5, OCTOBER 2012

Let the sum of deection angles N 1 i = , i.e., the included i=1 angle between the Y -axis and the reverse extension line of the N th link. Then we can obtain the line passing (xN , yN ) and perpendicular to the N th link as follows:

y = (x xN ) tan( ) + yN = yN (x xN ) tan . Let y = 0, and we can get the intersection point yN + xN x= (8) tan that corresponds to the turning radius. By substituting (7) into (8), we can obtain the turning radius as follows: R= L1 +
N i=2

Li cos( tan

i1 j =1

j )

i1

+
i=2

Li sin
j =1

j
Fig. 8. Conguration of the formulated CPG model. (a) Simplied structure. (b) CPG network conguration.

(9) where L1 , L2 , . . . , LN represent the mechanical dimensions of the robot and only 1 , 2 , . . . , N are undecided variables. Different turning radii will be obtained by applying different deection angles of modular units. Moreover, the overall deection angle will fundamentally determine the turning radius that will be xed next. Obviously, different parameter sets {1 ,2 , . . . , N 1 } will generate different turning trajectories and radii, which further need to be optimized. A more general form of (9) can be rewritten as follows: R(1 , 2 , . . . , N 1 ) = +
i=2

arbitrary modular units. More importantly, it will offer signicant benets to the design and locomotion control of modular amphibious robots in the future. IV. UNDERWATER LOCOMOTION CONTROL As for underwater locomotion, sh- or dolphin-like swimming is the primary locomotion means for the AmphiRobotII. A large number of methods have been proposed to tackle this issue [18]. At large, control methods employed can be categorized into two fundamentally different classes: trajectory tracking control and online gait generation control. In the former, predened swimming modes are usually generated via ofine planning and online tracking control. A good example is a method to replicate swimming kinematics through sh body wave tting [19]. In the latter, swimming gaits are not predened in advance, but calculated online. Inspired by lampreys, whose fast axial undulations being propagated as traveling waves are governed by activities in its spinal neural network named CPGs, more recent studies use CPGs to generate the desired gaits online for modular designs [20]. In essence, the CPGs are networks of neurons that can produce coordinated oscillatory signals without oscillatory inputs from sensory feedback or from higher control centers [21]. More specically, compared with the sine-based approach commonly used in sh-like swimming control, CPGbased control has the advantage of online gait generation and smooth trajectory transition even with relatively simple control signals [22], [23]. In this paper, an improved CPG model is responsible for generating the waves of joint activation for multimodal sh- or dolphin-like swimming. As shown in Fig. 8, a conceived CPG model functionally matching the mechanical system of the robot is created. It comprises a pair of pectoral CPGs and a series of tail CPGs. For the sake of simplicity and modularity, the pectoral CPG and the tail CPG possess the same structures, which are activated by the brain and regulated by feedback information. As the cell of the CPG model, a phase oscillator with controlled amplitude

L1 +
N

N i=2

Li cos( tan
i1

i1 j =1

j )

Li sin
j =1

j
N 1

LN L1 + + Li tan sin i=2 i1 1 cos j + sin tan j =1 = = +

i=2

i1

j =1

LN L1 + tan sin
N 1

Li cos j . sin j =1

i1

(10)

It is apparent that the minimum turning radius can be obtained by setting i = 0 (i = 1, 2, . . . , N 2), N 1 = , i.e., the minimum turning radius corresponds to the deection of the last modular unit separately. So the minimum turning radius can be derived as follows: LN Li + tan sin i=2 i=1 (11) that conforms to R in (5) by substituting D, D3 , D4 , and for L2 , L1 , L3 , and individually (N = 3). As demonstrated in (11), the body-deformation steering approach can be expanded to a similar locomotor conguration like our AmphiRobot with R= Li cos = sin LN L1 + + tan sin
N 1 N 1

YU et al.: ON A BIO-INSPIRED AMPHIBIOUS ROBOT CAPABLE OF MULTIMODAL MOTION

853

in terms of the Kuramoto model is utilized to achieve the synchronization of multiple oscillators so as to ensure coordinated movements of multiple joints [22], [24] i = 2fi + j T (i) aj wij sin(j i ij ) (12) a = i 4i (Ai ai ) ai i i = ai { 1 + cos(i )}. For detailed description of the oscillator model, we refer the readers to [25]. We just remark here that i for oscillator i represents the resultant burst serving as the output signal. As observed in Fig. 8(b), the tail CPGs include eight oscillators, i.e., O1O8. Every two oscillators constitute a CPG unit for each oscillating joint (J1J4), corresponding to a set of mutually inhibited extensor and exor. The subtraction of output of the left and right oscillators in each oscillating joint is utilized to actuate the corresponding servomotor, that is, i = i+4 i (i = 1, . . . , 4). For the pectoral CPG O9 or O10, its output signal directly serves as the control command to drive the pectoral n. The actuated joint angle signals of the pectoral ns can be computed as L = 9 a9 , R = 10 a10 , which have been modulated to be positivenegative. In order to establish the CPG-based control architecture, a bioinspired saturation function is introduced to identify the impact of the input drive on the output of the CPG. Compared with the previous research [22], a new saturation function building a bridge between input drive and oscillator response is presented. Going back to nature, most of the time, the carp (cyprinoid) devotes part of its tail or caudal n to propulsion and maintains slow cruising. The whole exible body is only involved when speeding up or in high-speed swimming with a higher oscillatory frequency [26]. Loosely inspired from the speed regulation observed in cyprinoids, an active oscillatory length-based control strategy is used to achieve that different lengths of the body participate in the oscillations at various speeds. Specically, two piecewise frequency and amplitude saturation functions are dened as follows: fi = gf (d) = Ai = gA (d) = kf ,i d + bf ,i flow cut kA ,i d + bA ,i Alow cut dlow ,i d dhigh 0 d < dlow ,i (13)

TABLE II LOCOMOTOR PERFORMANCE COMPARISON

which is denoted by dlow ,p ec . Hence, such a exible control policy will endow the robot with energy-efcient swimming. Due to the complexity, nonlinearity, and strong coupling underlying the CPG models, analytical methods have limited effect on parameter determination. Meanwhile, optimization or learning-based parameter selection is time consuming and requires self-contained sensory devices as a support for online sensing and evaluation. In this paper, initial testing of the CPG model is conducted in a dynamic simulation environment for solving swimming performance and parameter explorations [27]. Subsequent experiments on the physical robot are done to rene the preliminary parameter space till a satisfactory performance resulted. Notice that nonrhythmic turning can be induced when asymmetrical drives are applied to the left and right sides of the CPG model. The robot will then turn toward the side receiving the higher drive. Besides the asymmetrical drives inducing turning, the differences in oscillatory frequencies and amplitudes of the pectoral ns on bilateral sides will also assist in turning and stable propulsion. V. EXPERIMENTS AND DISCUSSION A. Mechanism Test In contrast to AmphiRobot-I, as summarized in Table II, the improved AmphiRobot-II shows an enhanced performance under testing. Waterproof tests demonstrate the robot can continuously operate about 1.5-h underwater without bearing the risk of water seepage. This will be of great assistance for extensive underwater experiments. The updated wheel-propeller-n mechanism facilitates rapid switching of actuators and locomotor patterns when adapted to varying environments. As shown in Fig. 9, the robot can adapt to various terrains successfully. B. Terrestrial Test To verify the proposed body-deformation steering approach, extensive experiments have been conducted in both eld and laboratory environment. In terms of the aforementioned principle, with the head and the rst two propelling units maintaining a straight line during testing, the third unit departs from the central position by a specic angle of 37 that remains invariable throughout. The achieved circular motion on the ceramic tile-paved oor is depicted in Fig. 10. The measured turning radius is about 550 mm, whereas the calculated radii from (5) is 552 mm. The minor error between the theoretical and experimental performances demonstrates the validity of the proposed body-deformation steering approach.

dlow ,i d dhigh (14) 0 d < dlow ,i

where d indicates the input drive signal received by the robot, which can further be divided into the left and right drive signals dL and dR associated with bilateral neural oscillators. flow cut and Alow cut represent the low-cut frequency and amplitude when oscillations block, respectively. kf ,i , bf ,i , kA ,i , bA ,i stand for frequency coefcients and amplitude coefcients, which decide the evolution of the intrinsic frequency and amplitude of the ith oscillator. dlow ,i and dhigh are the lower and upper thresholds separately. For the ith oscillator, once the drive reaches the corresponding lower threshold dlow ,i , the homologous oscillations start. By regulating the lower threshold dlow ,i , the corresponding joint can be enabled or not (e.g., dlow ,i > dhigh ), which can implement various propulsion modes coordinated by multiple propulsive elements. These are the same to the pectoral ns,

854

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO. 5, OCTOBER 2012

Fig. 9. Snapshots of moving on different terrains involving: (a) bumpy ground; (b) stone road; (c) soft grassland; and (d) slope.

Fig. 11. Relationship between drive and forward speed. Note that each experimental data point is an average of ve repeated trials, and that each data are not computed from the total distance traveled, but from the distance between the starting and the end points after a certain amount of time. TABLE III TURNING RADII WITH DIFFERENT INPUT DRIVES

Fig. 10. Image sequence of performing circular motion via the bodydeformation steering. Note that the blue and red lines in two snapshots represent the same locations in the actual ground.

C. Underwater Test To validate the feasibility of the proposed swimming control methods, extensive experiments are carried out in the swim pool and lake. By applying the same drive between the left and right sides of the CPG model, i.e., dL = dR , the AmphiRobot will perform forward swimming. A general trend is found that the forward speed increases directly with the increasing drive. Interestingly, as shown in Fig. 11, the overall prole is evidently divided into two distinct phases by point dlow ,1 . In the rst stage, besides the increase of frequency and amplitude, the active body length involved in the swimming rises continually as a result of the participation of more joints. Nevertheless, all the joints function and the active body length maintain invariant in the second stage. The two-phase prole demonstrates that the oscillating body length predominantly plays an important role in the propulsion of the AmphiRobot, and that the created saturation function has been effective in this regard. Meanwhile, the inuence of the drive difference on the turning radius is explored. Table III displays four cases of which dL and dR vary in the opposite direction while their means remain constant. As can be clearly observed, the drive difference between dL and dR can greatly impact the turning radius. The bigger the difference is, the smaller the turning radius will be, which is consistent with the biological fact [28]. Aside from sh-like propulsion, other locomotive patterns such as dolphin-like swimming, propeller mode, and landwater transition are examined. On the one hand, when the swivel mechanism is rotated 90 , the CPG model, without any further modi-

cation, will successfully take over the dolphin-like swimming. On the other hand, by exploiting the integrated sensory feedback, autonomous locomotion indicating interaction between the CPG-based control system and the external stimuli becomes possible. For instance, autonomous landwater transition can be elicited by sensory information gathered from onboard liquidlevel sensors. We remark that, to rapidly identify the environment, two liquid level sensors are mounted on the bottom of the head and the last module adjacent to the peduncle, respectively. Only when the last propelling element enters the water, wherein the rear liquid-level sensor detects the liquid state, will the robot switch from crawling gait to the swimming mode, and vice versa. Fig. 12 shows some on-site experimental scenarios on multimodal motions, which achieve the goal of ensuring our built propulsive mechanisms a relative satisfactory amphibious function. D. Discussion The built amphibious robot, with the aid of the proposed design scheme and control methods, exhibits a fairly good performance on land. However, when performing

YU et al.: ON A BIO-INSPIRED AMPHIBIOUS ROBOT CAPABLE OF MULTIMODAL MOTION

855

been done. For the terrestrial locomotion, a body-deformation steering method has been proposed with the minimum turning radius obtained accordingly; moreover, this method is extended to deal with a modular amphibious robot with N joints. For the underwater locomotion, a CPG-centered control model is employed, in which bioinspired saturation functions are proposed to accomplish an active oscillatory length-based control strategy. Physical experiments further verify the effectiveness of the proposed design. As a bonus, multimodal amphibious motions have been realized via modulating the CPG parameters or the switching mechanism. Future research will concentrate on two aspects: rst, great mechatronic efforts should be devoted to the development of enhanced amphibious mechanisms; second, parameter tuning and gait optimization by using some optimization techniques (e.g., [29], [30]) will be explored in depth. ACKNOWLEDGMENT The authors would like to thank the anonymous reviewers and the Associate Editor for their valuable comments and suggestions on improving the manuscript.
Fig. 12. Snapshots of underwater test. (a) Fish-like forward swimming. (b) Turning maneuver. (c) Switch between sh-like and dolphin-like swimming. (d) Propeller mode. (e) Locomotion transition from land to water.

REFERENCES
[1] L. Zollo, S. Roccella, E. Guglielmelli, M. C. Carrozza, and P. Dario, Biomechatronic design and control of an anthropomorphic articial hand for prosthetic and robotic applications, IEEE/ASME Trans. Mechatronics, vol. 12, no. 4, pp. 418429, Aug. 2007. [2] Z. Chen, S. Shatara, and X. Tan, Modeling of biomimetic robotic sh propelled by an ionic polymer-metal composite caudal n, IEEE/ASME Trans. Mechatronics, vol. 15, no. 3, pp. 448459, Jun. 2010. [3] B. Mazzolai, A. Mondini, P. Corradi, C. Laschi, V. Mattoli, E. Sinibaldi, and P. Dario, A miniaturized mechatronic system inspired by plant roots for soil exploration, IEEE/ASME Trans. Mechatronics, vol. 16, no. 2, pp. 201212, Apr. 2011. [4] H. S. Park and M. Sitti, Compliant footpad design analysis for a bioinspired quadruped amphibious robot, in Proc. IEEE/RSJ Int. Conf. Intell. Robot. Syst., St. Louis, MO, Oct. 2009, pp. 645651. [5] Z. Zuo, Z. Wang, B. Li, and S. Ma, Serpentine locomotion of a snakelike robot in water environment, in Proc. IEEE Int. Conf. Robot. Biomim., Bangkok, Thailand, Feb. 2009, pp. 2530. [6] L. A. Hirano, L. S. Martins-Filho, R. O. Duarte, and J. F. de Paiva, Development of an amphibious robotic propulsor based on electroactive polymers, in Proc. 4th Int. Conf. Auton. Robot. Agents, Wellington, New Zealand, Feb. 2009, pp. 284289. [7] A. S. Boxerbaum, P. Werk, R. D. Quinn, and R. Vaidyanathan, Design of an autonomous amphibious robot for surf zone operation: Part I mechanical design for multi-mode mobility, in Proc. IEEE/ASME Int. Conf. Adv. Intell. Mechatronics, Monterey, CA, Jul. 2005, pp. 14591464. [8] J. Ayers, Underwater walking, Arthopod Struct. Dev., vol. 33, pp. 347 360, 2004. [9] H. Greiner, A. Shectman, C. Won, D. Elsley, and P. Beith, Autonomous legged underwater vehicles for near land warfare, in Proc. Symp. Autonom. Underwater Vehicle Technol., Monterey, CA, Jun.1996, pp. 4148. [10] H. Yamada, S. Chigisaki, M. Mori, K. Takita, K. Ogami, and S. Hirose, Development of Amphibious Snake-like Robot ACM-R5, in Proc. 36th Int. Symp. Robot., Tokyo, Japan, Nov. 29/Dec. 1, 2005, pp. 433440. [11] A. Crespi, A. Badertscher, A. Guignard, and A. J. Ijspeert, Swimming and crawling with an amphibious snake robot, in Proc. IEEE Int. Conf. Robot. Autom., Barcelona, Spain, Apr. 2005, pp. 30243028. [12] Q. Yang, J. Yu, M. Tan, and W. Wang, Preliminary development of a biomimetic amphibious robot capable of multi-mode motion, in Proc. IEEE Int. Conf. Robot. Biomim., Sanya, China, Dec. 2007, pp. 769774. [13] W. Wang, J. Yu, R. Ding, and M. Tan, Bio-inspired design and realization of a novel multimode amphibious robot, in Proc. IEEE Int. Conf. Autom. Logist., Shenyang, China, Aug. 2009, pp. 140145.

body-deformation-based steering, frequent lateral displacement of the third propelling unit will impose a detrimental effect on the mechanism. To remedy this, the reciprocally rotating joint should be well lubricated to eliminate any damping during long-time operations. For the sake of long-term durability, an integrated design methodology for lightweight materials and smooth structures should be further investigated. As for the CPG-centered swimming control, many questions remain unsolved in the interplay between CPG modeling and onboard implementation. For instance, how to combine rhythmic and nonrhythmic motions into a general control framework, how to possibly simplify the articial CPGs system yet with the advantages of biological CPGs, and the like, such bottlenecks might limit extensive applications of the CPG architecture as well as robotic performance to some extent. More cooperative effort should be devoted to developing a more generalized CPG control framework. Notice also that one main drawback of the developed amphibious robot is relatively poor climbing ability on the slope due to insufcient torque supplied by the dc motors of the wheel-propeller ns. For this reason, more powerful dc motors with adequate torque-to-mass ratios will lead to a better multimodal locomotion performance both on land and in water. VI. CONCLUSION AND FUTURE WORK In this paper, we have presented a novel amphibious robot with the main emphasis placed on its propulsive mechanism and locomotion control. According to the structure and locomotion characteristics of the conceived amphibious robot, a series of modications of the morphological mimicry, the wheelpropeller-n mechanism as well as the swivel mechanism have

856

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 17, NO. 5, OCTOBER 2012

[14] M. Sfakiotakis, D. M. Lane, and J. B. C. Davies, Review of sh swimming modes for aquatic locomotion, IEEE J. Ocean. Eng., vol. 24, no. 2, pp. 237252, Apr. 1999. [15] C. Grand, F. Benamar, and F. Plumet, Motion kinematics analysis of wheeled-legged rover over 3D surface with posture adaptation, Mech. Mach. Theory, vol. 45, pp. 477495, 2010. [16] R. J. Bachmann, F. J. Boria, R. Vaidyanathan, P. G. Ifju, and R. D. Quinn, A biologically inspired micro-vehicle capable of aerial and terrestrial locomotion, Mech. Mach. Theory, vol. 44, pp. 513526, 2009. [17] K. Kikuchi, K. Sakaguchi, T. Sudo, N. Bushida, Y. Chiba, and Y. Asai, A study on a wheel-based stair-climbing robot with a hopping mechanism, Mech. Syst. Signal Process., vol. 22, pp. 13161326, 2008. [18] J. E. Colgate and K. M. Lynch, Mechanics and control of swimming: A review, IEEE J. Ocean. Eng., vol. 29, no. 3, pp. 660673, Jul. 2004. [19] J. Yu, M. Tan, S. Wang, and E. Chen, Development of a biomimetic robotic sh and its control algorithm, IEEE Trans. Syst., Man, Cybern. B, Cybern., vol. 34, no. 4, pp. 17981810, Aug. 2004. [20] A. Kamimura, H. Kurokawa, E. Yoshida, S. Murata, K. Tomita, and S. Kokaji, Automatic locomotion design and experiments for a modular robotic system, IEEE/ASME Trans. Mechatronics, vol. 10, no. 3, pp. 314325, Jun. 2005. [21] A. J. Ijspeert, Central pattern generators for locomotion in animals and robots: A review, Neural Netw., vol. 21, no. 4, pp. 642653, 2008. [22] A. J. Ijspeert, A. Crespi, D. Ryczko, and J.-M. Cabelguen, From swimming to walking with a salamander robot driven by a spinal cord model, Science, vol. 315, no. 5817, pp. 14161420, 2007. [23] C. Zhou and K. H. Low, Kinematic modeling framework for biomimetic undulatory n motion based on coupled nonlinear oscillators, in Proc. IEEE/RSJ Int. Conf. Intell. Robot Syst., Taiwan, Oct. 2010, pp. 934939. [24] J. A. Acebron, L. L. Bonilla, C. J. Perez, F. Ritort, and R. Spigler, The Kuramoto model: A simple paradigm for synchronization phenomena, Rev. Mod. Phys., vol. 77, pp. 137185, 2005. [25] R. Ding, J. Yu, Q. Yang, M. Tan, and J. Zhang, Robust gait control in biomimetic amphibious robot using central pattern generator, in Proc. IEEE/RSJ Int. Conf. Intell. Robot Syst., Taipei, Taiwan, 2010, pp. 3067 3072. [26] G. Wu, Y. Yang, and L. Zeng, Kinematics, hydrodynamics and energetic advantages of burst-and-coast swimming of koi carps (Cyprinus carpio koi), J. Exp. Biol., vol. 210, pp. 21812191, 2007. [27] R. Ding, J. Yu, Q. Yang, M. Tan, and J. Zhang, CPG-based dynamics modeling and simulation for a biomimetic amphibious robot, in Proc. IEEE Int. Conf. Robot. Biomim., Guilin, China, 2009, pp. 16571662. [28] F. A. Mussa-Ivaldi and S. A. Solla, Neural primitives for motion control, IEEE J. Ocean. Eng., vol. 29, no. 3, pp. 640650, Jul. 2004. [29] N. Hansen and A. Ostermeier, Completely derandomized self-adaptation in evolution strategies, Evol. Comput., vol. 9, no. 2, pp. 159195, 2001. [30] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, A fast and elitist multiobjective genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput., vol. 6, no. 2, pp. 182197, Apr. 2002.

Rui Ding received the B.E. degree in automation from Wuhan University, Wuhan, China, in 2006. He is currently working toward the Ph.D. degree in control theory and control engineering at the Institute of Automation, Chinese Academy of Sciences, Beijing, China. His research interests include mechatronics, biomimetic robotics, and embedded system.

Qinghai Yang received the B.E. degree in measurement and control and the M.E. degree in measuring and testing technologies and instruments from Hunan University, Changsha, China, in 2002 and 2005, respectively, and the Ph.D. degree in control theory and control engineering from the Institute of Automation, Chinese Academy of Sciences, Beijing, China, in 2009. He is currently an Engineer with the Research Institute of Petroleum Exploration and Development, Petrochina, Beijing. His research interests include biomimetics and automation.

Min Tan received the B.Sc. degree from TsingHua University, Beijing, China, in 1986, and the Ph.D. degree from the Institute of Automation, Chinese Academy of Sciences (IACAS), Beijing, China, in 1990, both in control science and engineering. He is currently a Professor with the Laboratory of Complex Systems and Intelligence Science, IACAS. He has published over 100 papers in journals, books, and conferences proceedings. His research interests include robotics and intelligent control systems.

Weibing Wang received the B.E. degree from Xinjiang Institute of Technology, Urumchi, China, in 1991. He graduated from the Department of Agricultural Mechanization, Zhejiang University, Hangzhou, China, in 2000. He is currently an Associate Professor with the Machine and Electricity Engineering College, Shihezi University, Xinjiang, China. His research interests include mechanical CAD/CAM and robotic technologies.

Junzhi Yu received the B.E. degree in safety engineering and the M.E. degree in precision instruments and mechanology from the North China Institute of Technology, Taiyuan, China, in 1998 and 2001, respectively, and the Ph.D. degree in control theory and control engineering from the Institute of Automation, Chinese Academy of Sciences (IACAS), Beijing, China, in 2003. After graduation, he was a Postdoctoral Researcher with the Center for Systems and Control, Peking University, Beijing. From March to August 2008, he was a Research Fellow with the City University of Hong Kong, Kowloon, Hong Kong. From September 2009 to August 2010, he was a Guest Researcher supported by the Alexander von Humboldt Foundation with the Department of Informatics, University of Hamburg, Hamburg, Germany. He is currently an Associate Professor with the Laboratory of Complex Systems and Intelligence Science, IACAS. He has published over 100 journal and conference papers. His research interests include biomimetic robots, multirobot systems, and intelligent information processing. Dr. Yu serves as an Associate Editor of the Journal of Mechanical Science and Technology as well as the International Journal of Information and Systems Sciences.

Jianwei Zhang received the Bachelor of Engineering (with distinction) and Master of Engineering degrees from the Department of Computer Science of Tsinghua University, Beijing, China, in 1986 and 1989, respectively, and the Ph.D. degree from the Department of Computer Science, Institute of Real-Time Computer Systems and Robotics, University of Karlsruhe, Karlsruhe, Germany, in 1994. He is currently a Professor and the Head of Technical Aspects of Multimodal Systems, Department of Informatics, University of Hamburg, Hamburg, Germany. His research interests include multimodal information systems, novel sensing devices, cognitive robotics, and humancomputer communication. He has published over 200 journal and conference papers, technical reports, four book chapters, and two research monographs. Dr. Zhang is a recipient of several awards, including the IEEE ROMAN Best Paper Award in 2002 and the IEEE AIM Best Paper Award 2008. He is a member of the organizing committees of numerous international conferences, including some future ones such as IEEE ICRA 2011 Program Cochair, IEEE MFI 2012 General Chair, IROS 2015 General Chair, etc.