1. Introduction
Humanity has found the most effective way to move on land with the invention of the wheel. However, the need to move in different environments where wheeled vehicles cannot move forces scientists to develop mechanisms that mimic living things. Although a very small part of the Earth’s surface is suitable for the locomotion of wheeled vehicles, living things can reach almost all surfaces of the world due to their unique physiology. Today, robots can perform different locomotion movements, such as walking, running, jumping, and swimming, that mimic living things. However, today’s robots can perform these movements by consuming a large amount of energy, while living things can perform more complex and stable locomotion using much less energy. Therefore, it is important to understand the principles of animal locomotion to develop robots with low energy consumption. Research on this subject has focused on the energy consumed during locomotion [
1,
2,
3,
4,
5]. The studies on the locomotion of legged robots have been mainly in the form of detailed examination of the movements of animals and the physical models related to these movements [
6].
Legged robots can be divided into three main families. There are robots equipped with rigid limbs, whose locomotion is essentially based on rigid body dynamics. They have been extensively studied since the last decades of the last century [
7,
8,
9,
10]. There are hopping robots that mimic hopping locomotion used by many animals, especially at high speeds. Hopping locomotion requires the presence of elastic limbs able to store elastic energy [
10,
11,
12]. Finally, there are soft robots, which are the most recent family. They have some characteristics of the hopping robots, but they also mimic the behavior of many animals that adjust the stiffness and shape of their limbs in order to increase their mobility [
13,
14,
15,
16].
The first issue of this research is vibration-based locomotion. Since this kind of locomotion shares many features and problems with hopping locomotion, a brief literature review on this topic is presented. The first research on hopping robots was carried out in the 1980s at the Leg Lab of the MIT [
10]. As a result of these studies, robot mechanisms capable of two-dimensional and three-dimensional locomotion have been developed, but these mechanisms had complex structures and showed excessive energy consumption. In contrast, appropriately designed elastic limb mechanisms allow the development of legged robots with lower energy consumption by exploiting a natural vibration, and the energy efficiency of these systems may approach one of biological organisms [
17]. The use of elastic limbs in robots brings energy efficiency together with a range of control difficulties. This difficulty can be eliminated by overlapping frequencies of the free vibrations of the elastic limbs and the periodic actuation of the robot. The identification of the stiffness properties of hopping robots is very important in order to generate accurate dynamic models, and, for this purpose, Experimental Modal Analysis (EMA) can be used [
18,
19]. In recent years, many robotic systems with elastic limbs have been developed [
20,
21,
22,
23,
24,
25]. For example, Bhatti et al. have developed a simple but effective control unit for a single-legged hydraulic robot, which can change the jump height, step length, and, thus, the flight time [
21]. Li et al. presented a soft-bodied jumping robot called JelloCube, which can be used for applications that require locomotion on uneven terrain [
22]. Using a free vibration mode, Reis et al. have developed an elastic mechanism that can perform locomotion with low energy consumption [
23,
24,
25].
The second main issue of this research is amphibious locomotion [
26,
27,
28,
29,
30]. The locomotion mechanism of many amphibious robots has been developed by taking inspiration from living things. The nervous system of amphibious animals has been studied, and a robot motion control system that can control movement patterns such as walking, breathing, and swimming has been developed and applied to a multi-foot amphibious robot [
31]. In Reference [
32], an amphibious spherical robot inspired by amphibious turtles was developed. The six-pedal amphibious robot [
33], which is a four-legged salamandra robot [
34] inspired by a cockroach, and the six-legged amphibious robot (AmphiHex-I) [
35] are some of the most well-known examples of a robot mimicking nature. Zhong et al. introduced an amphibious robot mechanism (AmphiHex-II) with semi-circular stiffness adjustable feet [
36]. In Reference [
37], an alligator-inspired robot was developed. Liquids have surface tension due to the strong cohesive force between their molecules. The surface tension of water is about ten times greater than the weight of some aquatic organisms with a body mass of milligrams. This makes it possible for aquatic insects to move on the water surface, and even to jump on it. Using this characteristic of the water, Koh et al. [
38] developed a water strider robot. In the robot design, a simple mechanical model of the interaction between the water strider’s legs and the water surface is established. Moreover, Yang et al. [
39] developed a theoretical leg motion model that provides an estimate of conditions for optimal jump performance of water runners.
This research deals with a simple two-legged robot, which consists of a U-shaped elastic beam, two feet, and a micro-DC motor with eccentric mass. Due to the shape and material of the feet, the robot can move both on land and in the water. Potential applications include the use of this robot to transport environmental sensors from ground to water (e.g., sensors for bridge monitoring in lagoons), and the collection and transport of plastic waste from shallow water to ground. This paper chiefly focuses on the aquatic locomotion of the amphibious robot. First, in order to have a better understanding of the locomotion principle, the natural frequencies and the modes of vibration of the amphibious robot are analytically derived. Since the terrestrial locomotion capabilities of this kind of robot have been already demonstrated in References [
23,
24,
25], the aim of the analysis is to demonstrate that, in the aquatic locomotion, the structural stiffness of the robot can be neglected. Then, a multi-body model of the robot is developed, in order to study the aquatic locomotion principle in detail. Many numerical results dealing with forces and displacements of the robot feet and locomotion speed are presented. Finally, experimental tests are described. The experimental results show that the robot has both terrestrial and aquatic locomotion capabilities, owing to the developed foot structure. The measured locomotion speed as a function of the frequency of the rotating mass and the natural frequencies of the robot are in good agreement with the results coming from the mathematical and numerical models.
4. Aquatic Locomotion Model
In this section, a multibody numerical model of the amphibious robot during locomotion in water is presented and discussed. The dynamic model is developed using Working Model (WM) 2D, and the numerical and experimental locomotion speed for different angular velocities of the rotating mass will be compared. This model will help to understand and discuss the locomotion principle in the water.
The WM model, which has been built using the geometrical and inertial parameters of
Table 1, and the interaction forces of feet with the water are presented in
Figure 5. A planar motion of the robot is assumed, and the frame of the robot is modeled as a rigid body, which is reasonable due to the high frequencies of the structural modes, the low interaction forces with the water, and the low locomotion speed.
The horizontal force on the front and hind foot (
and
, respectively) is mainly due to drag, which is related to the front and hind foot speed along the
-axis (
and
, respectively), but there is also a small lift force that arises when the foot profile moves vertically in the water. The following equations hold:
In Equations (38) and (39), is the density of water, and are the drag coefficients in the forward and backward directions along the -axis, respectively, and and are the vertical positions of the front and hind foot, so that and are the areas (subject to drag forces) of the submerged part of the front and hind foot, respectively. In Equation (40), and are the lift forces on the front and hind foot, respectively, is the lift coefficient of the foot subject to a vertical flow, and are the ratio of the submerged length of the front and hind foot with respect to the total foot length (2), respectively, and and are the front and hind foot speed along the -axis, respectively. As indicated in Equations (38) and (39), the drag forces are always opposite of the foot speed (usually , i.e., the foot is partially submerged). As indicated in Equations (38) and (39), the horizontal force is null if the foot is completely out of water.
The vertical force on the front and hind foot (
and
, respectively) is composed of two terms:
where
and
are the buoyancy forces on the front and hind foot, respectively, and
and
are the vertical drag forces on the front and hind foot, respectively.
The buoyancy forces on each foot can be modeled as in
Section 3.2:
where
is the acceleration of gravity, and
is the width of an equivalent rectangular cuboid with the same length (
), height (
), and volume (
) of the real foot. This is an approximation that allows to model the buoyancy force on each foot as a linear function of
(or
). Equations (42) and (43) show that the buoyancy force is null if the foot is completely out of the water.
The vertical drag force on the front and hind foot, which is due to the front and hind foot speed along the
-axis (
and
, respectively), can be modeled as:
where
and
are the drag coefficients in the forward and backward directions along the
-axis, respectively, and
is the equivalent area of the cross section (orthogonal to the
-axis) of the submerged part of the foot. As indicated in Equations (44) and (45), the drag forces always have the opposite sign with respect to the foot speed, and are null if the foot is completely out of the water.
Due to the shape of the foot,
and
have different values. In the dynamic simulations, it is assumed [
45,
46,
47]:
= 1.15,
= 2.15,
= 0.8,
= 0.8, and
= 0.045.
In
Figure 6, a sequence of screenshots is presented, which have been derived from a simulation performed with angular velocity of the rotating mass
= 24 rad/s (counter-clockwise). The configurations reported in
Figure 6 match well with the ones observed during the experimental tests. After an initial transient, the motion of the robot is periodic. Thus, the same motion (and the related sequence of configurations) is repeated every time period
. In
Figure 7, the forces on the front (
,
) and hind foot (
,
) as a function of time (
) are reported for a couple of cycles. It can be noticed that
and
have a very similar profile and are almost in phase. In addition,
and
have a similar profile, but
is shifted to about 0.065 s (which corresponds to a phase shift of about 90 deg) with respect to
, since the robot oscillates backward and forward (see
Figure 6) as it advances.
From the analysis of
Figure 6 and
Figure 7, the following considerations can be done. In the screenshot (a), the rotating link (carrying the eccentric mass) is vertical and pointing upwards and the feet of the robot are moving forward. In this configuration, the minimum (directed backwards and largest in modulus) horizontal drag forces are applied to the feet of the robot (see
and
at
= 0.04 s in
Figure 7). Moreover, in this configuration, the centrifugal force due to the rotating mass is pointing upwards, so the submerged part of the feet is less than in the static equilibrium configuration (rotating mass not moving).
In the screenshot (e) (which is a dual configuration with respect to (a)), the rotating link is vertical and pointing downward and the feet of the robot are moving backward. In this configuration, the maximum (positive, i.e., directed forward) horizontal drag forces are applied to the feet of the robot (see
and
at
= 0.17 s in
Figure 7). In this configuration, the centrifugal force due to the rotating mass is pointing downward, so the submerged part of the feet is more than in the static equilibrium configuration. Therefore, the submerged area of the feet subject to horizontal drag forces is higher than in (a), and this contributes to have a higher value (in modulus) of the forward force with respect to the backward force related to (a). Nevertheless, the main reason why the absolute value of the maximum forward force (for each foot) is higher than the absolute value of the minimum (backward) force is because
.
Similarly, in all the other configurations, the robot motion is dominated by the corresponding direction of the centrifugal force generated by the rotating mass. In particular, in configuration (b), the minimum value of
is reached, as it can be verified in
Figure 7 for
= 0.08 s and, indeed, the direction of the centrifugal force (i.e., the direction of the rotating link) is such that it reduces the vertical load on the front foot (and, thus, reduce the buoyancy force and
). In configuration (c), the maximum backward inclination of the robot (about −2.5 deg) is reached and the centrifugal force is horizontal so that the maximum destabilizing torque on the robot is achieved. In configuration (d), the maximum value of
is reached, as it can be verified in
Figure 7 for
= 0.14 s, and the direction of the centrifugal force (i.e., the direction of the rotating link) is such that it increases the vertical load on the hind foot (and, thus, increases the buoyancy force and
).
Exactly the same considerations can be done for configurations (f), (g), and (h), in which the centrifugal force makes the vertical load on the front foot increase (together with the related buoyancy force), makes the robot achieve its maximum forward inclination (about −2.5 deg), and makes (and the related buoyancy force) reach its minimum value, respectively.
It can be concluded that the main reason for robot net forward motion is that
due to the asymmetric shape of the feet. In the periodic forward-backward motion of the feet with respect to the water (due to the centrifugal force of the eccentric mass that pulls forward and backward the robot), a positive integral force due to horizontal drag forces allows the robot to advance. This can be easily verified looking at the profile of
and
in
Figure 7. The area of the positive part of the plot is higher than the area of the negative part of the plot for both feet. This phenomenon is similar to the one exploited in vibration conveying [
48], in which the inclination of the inertia force caused by vibrations increases the static friction force in a direction and decreases the static friction force in the opposite direction.
The dynamic behavior of the robot using a constant motor torque as an input has also been investigated. In particular, simulations have been carried out with torques (directed counterclockwise) from 0.00001 Nm to 0.01 Nm. For low torque values, i.e., from 0.00001 to 0.00004 Nm, the torque is not sufficient to rotate the eccentric mass. For torque values between 0.00004 and 0.0005 Nm, a stable periodic locomotion is obtained, after an initial transient. The locomotion speed, expressed in body length/mm (the body length is
= 85 mm), the motor speed, and their mean values (0.173 body length/s and 31 rad/s, respectively) are reported in
Figure 8 for an input torque of 0.00005 Nm.
For torque values higher than 0.00005 Nm, the rotating mass is continuously accelerated until the robot reaches a stable periodic locomotion with rotational speeds higher than 200 rad/s, which is not reasonable for this type of robot. For torque values higher than 0.01 Nm, the robot tips over.
It can be concluded that the use of a constant torque as an input allows for a stable locomotion only for motor torques between 0.00004 and 0.00005 Nm. The use of a constant angular velocity as an input is much more reasonable, since it allows us to excite the resonances of the system. Moreover, it is more flexible (since the locomotion speed can be precisely tuned), and it results in higher locomotion speeds, as detailed in
Section 5.
5. Experimental Results
During the experiments (see
video S1 published as
Supplementary Material), the angular velocity of the eccentric mass is controlled by the voltage applied to the DC motor. The aquatic and terrestrial locomotion of the robot is recorded with a high-speed camera while the robot is moving in a 0.7 × 1.5 m aquarium, and the locomotion speed, the rotating mass rotational speed, and the robot foot positions are measured thanks to high-speed camera recordings. The DC motor voltage is increased from 0.8 V to 2.7 V in 0.1 Volt increments.
In
Figure 9, the experimental and numerical aquatic locomotion speed as a function of
(counterclockwise rotation) are overlapped. The numerical curve has been derived using the WM model presented in
Section 4. It can be observed that there is good agreement between the resonance frequencies of the numerical and experimental model. A first resonance peak appears at about 26.5 rad/s and a second resonance peak appears at about 33–35 rad/s in both cases. These values are also in good agreement with the resonance frequencies computed with the linear model of
Section 3.2 (28.8 rad/s and 34.2 rad/s). Moreover, the numerical and experimental curves are in good agreement and nearly overlapped. For higher values of
(>40 rad/s), the numerical model cannot be used due to the increasing importance of the effect of waves generated by the robot (which are observed in the experiments). It is worth noting that the natural frequencies of aquatic locomotion are well below the natural frequencies of the structural modes studied in
Section 3.1. Hence, the assumption of rigid robot in the model of aquatic locomotion is consistent with the actual operation of the robot.
In
Figure 10, the displacements of the hind foot are presented both for terrestrial and aquatic locomotion, which have been measured in experiments with
= 38 rad/s. The experimental curves are almost linear, and this is related to constant locomotion speeds. In the case of aquatic locomotion, the experimental points are overlapped with the numerical curve, which is almost linear with small local peaks due to the periodic locomotion. The experimental and numerical results are in good agreement. A more detailed comparison between the experimental and numerical positions of the robot feet is depicted in
Figure 11 considering a couple of periods of motion. Experimental and numerical data are in good agreement in this time scale as well: the period of the oscillations is the same and the error between the numerical curve and the experimental points is always less than 1.3 mm.
The differences between numerical and experimental data in
Figure 9,
Figure 10 and
Figure 11 are mainly due to all the simplifying assumptions (detailed in
Section 4), the disturbance forces related to the electrical cables of the motor, and the effect of the small waves present in the experiment, which are not considered in the numerical model. The overlap between the experimental points and numerical curves could be improved by using computational fluid dynamics to simulate the flow around feet for different relative velocities and angles, or by using experimentally determined drag and lift coefficients. However, these activities are beyond the scope of the present work.
Figure 12 shows both the terrestrial and aquatic locomotion speed as a function of
, as measured in the experiments. The figure shows that the robot reaches the highest speed on the ground when the eccentric mass rotates at an angular velocity of about 38 rad/s, and it reaches the highest speed on the water when the eccentric mass rotates at an angular velocity of about 35 rad/s, which is in accordance with the resonance peak found with the mathematical and numerical models of
Section 3 and
Section 4. In
Figure 13, a sequence of photos showing an aquatic locomotion test with
= 25 rad/s is presented.
The proposed robot is able to carry out the transition ground-water and vice-versa as shown in
Figure 14 and in
video S1 (published as
Supplementary Material). Even if the ground locomotion is not the focus of the present work, it should be noted that the robot is very robust against big variations in the coefficients of friction (both static and dynamic) of the surface. In the transition water-ground, the feet are wet (which causes a significant decrease in the friction coefficients); nevertheless, the robot is able to successfully perform the transition. The investigation of the effect of friction coefficients in the ground locomotion and in the transition phases will be part of future work.