1

Design and Control of an Energy Accumulative

Hopping Robot
Samuel Burns, Matthew Woodward

Abstract—Jumping and hopping locomotion are efficient performance, and agility. A basic performance metric for
means of traversing unstructured rugged terrain with the former vertical jumping height, heightjump = energyjump /(mass ∗
being the focus of roboticists. This focus has led to significant gravity), is highly coupled to mass. To produce the highest
performance and understanding in jumping robots but with
limited practical applications as they require significant time performance jumping systems the energy storage part of the
arXiv:2312.08301v1 [cs.RO] 13 Dec 2023

between jumps to store energy, thus relegating jumping to a cycle can take tens of seconds, drastically limiting the robot’s
secondary role in locomotion. Hopping locomotion, however, can agility. Hopping robots, or robots that repetitively jump, pro-
preserve and transfer energy to subsequent hops without long vide a path forward as they are more efficient, recovering
energy storage periods. Therefore, hopping has the potential to energy from previous hops, and are more agile, as they do not
be far more energy efficient and agile than jumping. However, to
date, only a single untethered hopping robot exists with limited require long delays between hops to store energy, locomote,
payload and hopping heights (< 1 meter). This is due to the reposition, or stabilize; the initiation of hops is determined by
added design and control complexity inherent in the requirements the fundamental dynamic behavior as opposed to arbitrarily
to input energy during dynamic locomotion and control the by the controller.
orientation of the system throughout the hopping cycle, resulting Researchers have been studying hopping robots since they
in low energy input and control torques; a redevelopment from
basic principles is necessary to advance the capabilities of were first developed by Raibert et al. in 1984 [46], [47],
hopping robots. Here we report hopping robot design principles however, due to their design and control complexity from
for efficient and robust systems with high energy input and the necessity to maintain orientation and input energy over
control torques that are validated through analytical, simulation, the extremely short time periods between successive hops and
and experimental results. The resulting robot (MultiMo-MHR) complex ground interactions, four decades later and only an
can hop nearly 4 meters (> 6 times the current state-of-the-art);
and is only limited by the impact mechanics and not energy input. extremely small number have ever been developed with limited
The results also directly contradict a recent work that concluded payload, performance, and most being tethered [46]–[52].
hopping with aerodynamic energy input would be less efficient While hopping locomotion has been studied most notably in
than flight for hops greater than 0.4 meters. Salto, these robots do not change their locomotion energy (LE
Index Terms—Hopping, Robot, Control, Energy Accumulation, = sum of the desired kinetic, gravitational potential, and elastic
Efficiency potential energies) over successive hops, they maintain a
nearly constant level [50]. In addition, these robots do not have
sufficient performance to support a payload of task specific
I. I NTRODUCTION
components; with Salto’s average hopping height of 0.65 m

J UMPING robots have been under significant development

for decades demonstrating a diversity of designs, energy
storage and release mechanisms, and additional behaviors such
at 100 grams a 10 gram payload would significantly reduce
the system’s performance. Given that we expect systems with
energy input and loss to stabilize over time, we seek to un-
as gliding and self righting [1], [2]. Jumping locomotion tends derstand whether LE can not only be accumulated (increased)
to produce the highest specific energy (energy/mass) systems but done so through constant actuator power output. Nature
[3], [4], as compared to hopping because they do not require however provides an answer. Observations of desert kangaroo
the additional components necessary for hopping locomotion; rats (Dipodomys deserti) shows an extraordinary aspect of
namely those for orientation control. This has led to impres- their hopping locomotion; their LE increases over successive
sive, high specific energy, jumping robot designs with added hops. They accumulate energy with each cycle not only in
features demonstrating minimalist high performance jumping their forward kinetic energy but in their elastic tissues as
[3], [5], [6], micro jumping robots [7]–[11], linkage based well, all while their metabolic rate remains constant with
leg designs [12]–[15], linear leg designs [16]–[19], buckling increasing velocity [53]; imagine running as fast as you can
leg designs [20]–[22], spring leg design [3], self-righting but feeling like you are walking. These animals incredibly
[23]–[26], steering [13], [27], aerial orientation control [28], store, accumulate, convert, and release energy from previous
surface adaptation [29], gliding [30]–[36], flight [37], legged cycles to increase their LE over time without increasing their
[38]–[41], wheeled [15], [42], [43], and soft jumpers [44], averaged actuator power. Energy accumulation in robotics
[45]. However, jumping systems tend to recover no energy has been explored in a tethered one-degree-of-freedom robot,
from previous jumps and therefore have limited efficiency, however, the robot is only able to achieve an increase in
hopping height of approximately 35 mm as the dynamic
Manuscript received November 14, 2023; revised. behavior surpasses the actuator’s speed [54]. An additional
The authors are with the Robot Locomotion and Biomechanics Laboratory,
Mechanical Engineering Department, Tufts University, Medford, MA, 01890 work presented a simulation which shows an increase in
(email: [email protected]) hopping height of 250 mm, however, this behavior is not
 2

a Hopping Cycle c 0.15

90 0.2
Drop Stance Rebound

Body Mass Ratio (%)

0.3 MultiMo-MHR
80 0.4
Ed* EFDd Es* Emech Er* EFDr 0.5
TD TD - +
LO LO- +
70
0.6
Touchdown, ETD-+ Liftoff, ELO-+ 60 0.7
b αr* : 99% 90% 80% 70% 60% 50% 40% 50 0.8 Critical αr* (%)
14 0.85
Model 33.7 m @ 79 sec 40 -3
12 10 10-2 10-1 100
mB Total Robot Mass (kg)
Hopping Height (m)

10 d 90 0.125
MultiMo-MHR

Body Mass Ratio (%)

8 80 0.25 1
2
0.5
6 mF 70 0.75 3
4
60 5
4 6
50 7
2 Rebound Input Force (N)
8
40
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 5 Time (sec) 10 15 20 0.001 0.01 Total Robot Mass (kg)

Fig. 1. Energy accumulation in hopping robots. a) Hopping cycle: Phases (drop, stance, and rebound) and transitions (touchdown, liftoff). Normalized center-
of-mass energy modification (appendix: robot energy): Control inputs during the three phases (Ed∗ , Es∗ , Er∗ ), drag loss during drop (EF Dd ) and rebound
(EF Dr ), touchdown losses (ET D∓ ), liftoff losses (ELO∓ ), and mechanical losses (Emech ); where control inputs can be positive or negative and losses
are always negative. b). Energy accumulation in a simple hopping model with varying input ratios, αr∗ . c) Critical αr∗ for energy accumulation. b) Critical
input force, Fr∗ = αr∗ mT g, for energy accumulation. Note: b) Total mass, mT = mB + mF = 697 g, taken from the initial prototype. c-d) Added
crit
mass example. Red arrow indicates adding 100 g to the body mass and black arrow indicate adding the same mass to the foot. Note: b-d) The drag coefficient
and effective area is experimentally measured from the initial prototype and scaled by mass2/3 .

shown to exist in the robot [55]. The accumulation of energy at fundamentals of hopping locomotion, energy accumulation at
a constant actuator power output opens the possibility to high- constant actuator power output, efficiency, aerodynamic energy
performance hopping, as typically high performance jumping input, and control of a high-performance hopping robot.
and hopping requires proportionally larger actuators who’s
mass in turn reduces performance. Realizing this concept in a II. E NERGY ACCUMULATION C ONCEPT
robotic platform would not only eliminate the need for large Locomotion strategies tend to require cyclic motions in
actuators, but, through accumulation, allow a robot to achieve which the LE is modified over the course of the cycle. In
significant hopping heights. hopping locomotion, the hopping cycle can be divided into
With inspiration from Salto which utilizes propellers for three phases, including drop, stance, and rebound, and two
orientation control, its natural to consider using propellers transitions, including touchdown (TD) and liftoff (LO); as seen
for both orientation control and energy input for hopping. in Fig. 1a. Each phase has both the potential for a control input
Zhu et al. demonstrated this concept in the PogoDrone [37], to add or remove LE and a characteristic loss of LE, whereas,
however, as stated at 0.5 meters ”bouncing performs worse each transition has only a characteristic loss of LE; yielding
than simply hovering”; assuming the extra mass associated eight distinct energy modifications. The energy which must be
with the bouncing mechanism is removed, this system would input to overcome the losses and achieve a specific hopping
likely always perform better in hover. This poses a quandary, height ratio, δrd = hr /hd , is therefore,
as hopping and jumping are known to be efficient means Ed∗ + Es∗ + Er∗ = (δrd − 1)mT ghd
of overcoming obstacles in terrestrial locomotion and typi-
− EF Dd − ET D∓ − Emech − ELO∓ − EF Dr (1)
cally more efficient than powered flight [56]. However, as
stated, this robot is designed to periodically bounce to ”take where, total mass, mT , drop height, hd , rebound height,
samples of soil”, while the vast majority of time is spent hr , and gravity, g, are represented accordingly. The height
in flight; experimental results show thrust is never reduce ratio δrd > 1, δrd < 1, and δrd = 1 indicates increasing
below approximately 60% of hover, indicating a slowed decent (LE accumulation), decreasing (LE dissipation), and constant
which is not applicable to an efficient hopping behavior. The (constant LE) hopping heights; where the latter (constant LE)
conclusion of the work states that the robot may have the has been the focus of previous hopping robots. The control
”potential to efficiently operate in low heights”, but this is input energies during the drop, Ed∗ , stance, Es∗ , and rebound,
not explored. While this work has focused on the addition Er∗ , phases are specified by the controller. This accounts for
of a bouncing mechanism to a flying robot, we seek to all jumping and hopping robots as energy input can be through
explore the fundamental development of a high-performance aerodynamic forces (e.g., propellers, thrusters) that can add
hopping robot which utilizes a rotor configuration for energy LE throughout all phases (Ed∗ , Es∗ , Er∗ ) or leg forces (e.g.
input and orientation control. In particular, to understand the direct actuation, released elastic energy, combustion) that can
 3

only input LE during the stance phase (Es∗ ). Assuming an αr∗ = Er∗ /(mT ghr ), which is the ratio of the input force to
atmosphere, this highlights a significant benefit of aerody- the gravitational force; where, αr∗ ≥ 1 the robot is flying and
namic LE inputs, in addition to their potential dual use for αr∗ < 1 the robot is falling. Using the parameters of the initial
orientation control. Assuming an atmosphere, energy will be prototype and the determined drag and touchdown losses,
lost to aerodynamic drag (drop, EF Dd , and rebound, EF Dr ). Fig. 1b (equations 2, 3), shows energy can be accumulated
At touchdown, assuming high contact damping, the energy in without sufficient thrust for flight; where, even αr∗ = 40%
component(s) which make direct contact with the ground will produces an accumulation of energy over multiple hopping
be lost, leading to a touchdown loss, ET D∓ . The ∓ indicates cycles. Furthermore, as this force is only produced during
the transition from the instantaneous moment prior to contact the rebound phase, the force averaged over the full hopping
(-) to that after (+); where this simplifies the complex impact cycle would be even less; e.g., a quadrotor would have to
mechanics. During the stance phase the energy storage and continuously produce a force equal to the gravitational force,
release mechanisms move, creating loses due to friction in the however, a hopping robot producing less force, would only be
mechanisms and dissipation in the elastic material, leading required to do so over the rebound phase; this contradicts the
to a mechanical loss, Emech . At liftoff, as energy must be conclusions reached by the previous work [37]. The hopping
transferred between components such that the entire system robot therefore has the potential to be substantially more
is moving at a constant velocity, energy may be lost in the efficient and yield longer operational times than those of
transfer, leading to a liftoff loss, ELO∓ . The ∓ indicates the rotorcraft robots. However, assuming spring stiffness and mass
transition from the instantaneous moment prior to liftoff (- do not change, spring-mass systems have a constant natural
) to the moment afterP(+). The hopping cycle efficiency is frequency, that translates into a nearly constant ground contact
therefore, ηcyc = 1 − |Ei |/mT ghd , where, Ei ’s represent time across varying hopping heights. Therefore, accumulating
the energy losses; which may include required control inputs energy with direct leg actuation would fundamentally require
that remove energy (e.g., stabilization). an increase in actuator power output over the stance phase;
while average power may remain constant as the aerial phase
A. Energy Accumulation Design. time would increase.
The minimum point at which a hopping robot will accumu-
To explore the accumulation of energy, a simple model is
late energy with energy input during the rebound, is the critical
proposed, Fig. 1b. Assuming the main spring and the ground
αr∗ , Fig. 1c; where the critical input force, Fr∗ = αrcrit
∗ mT g,
contact are modelled as springs with dampers, the dynamics
is seen in Fig. 1d. Over four orders-of-magnitude in robot mass
are governed by,
and body mass percentages from 0.4 to 0.95, Fig. 1c shows
z̈B = (−mB g + UB + bB (żF − żB ) + kB (zF − zB ))/mB the critical αr∗ is highly dependent on body mass percentage,
(2) mB /mT , and nearly constant (slight negative slope) in total
z̈F = (−mF g + UF + bB (żB − żF ) + kB (zB − zF ) robot mass, mT . Therefore, as body mass increases the critical
αr∗ decreases, whereas, increases in foot mass significantly
− bF żF − kF ∗ zF )/mF (3)
increases the critical αr∗ . Assuming the actuator’s mass is
where, the vertical position, zB , mass, mB , spring constant, proportional to its output force, this indicates that body mass
kB , and damping coefficient, bB , of the body and the vertical increases will result in proportionally less actuator mass,
position, zF , mass, mF , spring constant, kF , and damping whereas, foot mass increases will result in proportionally
coefficient, bF , of the foot are represented accordingly. Fig- more actuator mass (i.e., more of the system’s mass will be
ure 1b,c assumes the body input force, UB , is the sum of devoted to the energy input actuators). Moreover, from Fig.
the drag force and, during the rebound, the control input 1d, the critical force, Fr∗ , may remain nearly constant with
force, Fr∗ = αr∗ mT g; the foot input force, UF = 0. A increases of body mass; however, this is dependent on the
conservative estimate of the drag coefficient and effective body mass ratio and total mass of the robot. Therefore, given
area, CD A = 0.072122, is assumed constant due to the low an existing system and a required body mass increase, the
slope of the linear fit from appendix: experimental energies. critical force can be use to determine the required increase in
From experimental observations, the foot-ground damping is actuator power output or reduction in foot mass necessary to
assumed to be high, while the body-foot damping is assumed maintain the same actuator power output. As an example, a
negligible during spring extension. In simulation, the spring 100 gram increase to the prototype’s (MultiMo-MHR) body
and damping value are modified to account for the current mass (red arrow) results in nearly no change in the required
operational phase and relative position of the components and thrust, whereas, the same increase in the foot mass (black
the ground. arrow) would require nearly twice the thrust. This creates a
Focusing on the addition of LE during the rebound phase, fundamentally new design paradigm for hopping robots, as
Er∗ , using aerodynamic forces, the rebound control inputs will larger body masses require proportionally less input force, as
assume a constant actuator power output and thus constant a function of the mass, to accumulate energy, therefore freeing
aerodynamic force (thrust). However, instead of energy con- up mass for additional components and complexity.
sumed by the actuator, we will consider work done on the The fundamental scaling relationship between energy modi-
robot. Therefore, the work done by the input force, Fr∗ , on fications can be determined by normalizing the changes to the
the robot is, Er∗ = Fr∗ hr . Normalizing Er∗ by the rebound LE by the characteristic energy of the phase or transition. At
energy, mT ghr , produces the rebound control input ratio, the initiation of the drop phase the robot has an initial energy
 4

a b -1 40 .0 50 60
1.0
30

10

20

5
1 70

0.
181 mm
-2 2 2.0 0.9

Ground Contact Vertical Velocity (m/s)

Bumper 4.0 80
0.8
-3

0
90

Liftoff Vertical Energy (%)

Elastic Energy Negative Vertical 0.7
Thrust -4
Storage Mechanism LO+ Velocity
Motor (x4) 4 0.6
Body -5
0.5
IMU 2 IMU 1 -6
0.4
Microcontroller
368 mm

-7 6
0.3
RF Transceiver Ground Contact
(on back) -8 0.2
Battery (x2) LO Angle (deg)
-9 LO+ Vertical Velocity (m/s) 0.1
Leg 8 Friction Coef. (min)
0
0 5 10 15 20 25 30 35 40 45
Ground Contact Angle (deg)
e Desktop: Robot:
Vicon Controllers (500 Hz):
Foot (Capture: 600 Hz) Position
(Transmit: 100 Hz) Position
X/Y/Z Control
dx Velocity PID
dt Mode Selection
Apex Detection Desired
• drop
-1 IMU 9 dof Attitude Roll/Pitch • stance
c d
Vertical Velocity (m/s)

-2 (200 Hz) • rebound

Roll/Pitch
Fusion
Ground Contact

-3 Gyroscope PID Roll/Pitch

-4 Attitude Attitude Control
3 dof Angular Rate
-5
Energy (%)

Energy (%)

Angular Rate
Rotational

Horizontal

42.7 96.3 (1000 Hz) Yaw

-6
Liftoff

Liftoff

PID Yaw Thrusters

-7
Accelerometer Control
-8
-9 3 dof
0 0 (1000 Hz)
Ground Contact Detection
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 45
Ground Contact Angle (deg) Ground Contact Angle (deg)

Fig. 2. Robot Design: a) MultiMo-MHR with combined monopedal hopping leg and quadrotor for control. b-d) Ground contact mechanics: Showing liftoff
(LO) angle, required friction coefficients, and translational and rotational energies. Note: LO+ indicates the moment after liftoff. e) Linear control scheme.

equal to mT ghd which is modified in eight distinct ways over The stance phase energy can be modified by the stance control
the course of the hopping cycle. Rewriting equation 1 leads input, Es∗ , and mechanical losses, Emech , resulting in the
to the the energy at the apex of the rebound phase of, energy just prior to liftoff of,

mT ghr =mT ghd + Ed∗ + EF Dd + ET D∓ + Es∗ 2 2

1/2mB vLO − = 1/2mB vT D − + Es∗ + Emech
+ Emech + ELO∓ + Er∗ + EF Dr . (4)
v2 −
1 + αs∗ + αmech = 2LO (7)
However, these energies are highly coupled making deter- vT D −
mination from experimental data and generalization difficult αs∗ = Es∗ /(1/2mB vT2 D− ), αmech = Emech /(1/2mB vT2 D− )
without separating the phases and transitions. During the drop,
the energy can be modify by the drop control input, Ed∗ , and
drag, EF Dd , resulting in the energy just prior to touchdown where, vLO− is the velocity just prior to liftoff. The robot then
of, undergoes the second transition in which energy can be lost
in the transfer between components, ELO∓ , resulting in the
1/2mT vT2 D− = mT ghd + Ed∗ + EF Dd energy just after liftoff of,
v2 −
1 + αd∗ + αF Dd = T D (5) 2 2
2ghd 1/2mT vLO + = 1/2mB vLO − + ELO∓ (8)
αd∗ = Ed∗ /(mT ghd ), αF Dd = EF Dd /(mT ghd )
where, vLO+ is the velocity just after liftoff. The robot then
where, vT D− is the velocity just prior to touchdown. The
enters the rebound phase in which the energy can be modified
robot then undergoes the touchdown transition, losing energy,
by the rebound control input, Er∗ , and drag, EF Dr , resulting
ET D∓ , resulting in the energy at the initiation of the stance
in the energy at the apex of the hop of,
phase,

1/2mB vT2 D− = 1/2mT vT2 D− + ET D∓ 2

mT ghr = 1/2mB vLO + + Er ∗ + EF Dr (9)
mB
1 + αT D∓ = (6)
mT where, hr is the rebound height and the drop point for the
αT D∓ = ET D∓ /(1/2mT vT2 D− ). subsequent hop. However, the velocity just after liftoff, vLO+ ,
 5

is difficult to measure. Therefore, to remove the velocity just TABLE I

after liftoff, equations 8,9 can be combined; leading to, ROBOT PARAMETERS

2 Parm. Description Value Qty

mT ghr = 1/2mB vLO − + ELO∓ + Er ∗ + EF Dr

mT ghr mT ghr mb Body mass 382.82 g 1

1 + αLO∓ + 2 (αr∗ + αF Dr ) = 2
mf Foot mass 102.48 g 1
1/2mB vLO − 1/2mB vLO − mbat Battery Mass 198.97 g 1
(10) ms Spring mass 12.27 g 1
mB mb + mbat + 1/2ms 587.93 g
αr∗ = Er∗ /(mT ghr ), αF Dr = EF Dr /(mT ghr ) mF mf + 1/2ms 108.62 g
2
αLO∓ = ELO∓ /(1/2mB vLO Robot Specific energy 23.87 J.kg −1
−) (11)
Combining these equations together yields the ratio of the
rebound height to drop height, δrd , as,
Assuming high damping in the foot-ground interface, the high
(αd∗ + ηF Dd )ηT D∓ (αs∗ + ηmech )ηLO∓ impact forces will be dissipated before the body can apply any
δrd = (12)
2 − αr∗ − ηF Dr significant forces. Therefore, the determining factor in required
where, the terms representing control inputs including, αd∗ , foot strength is the foot mass, foot-ground interface stiffness,
αs∗ , and αr∗ can be either positive or negative, indicating and touchdown velocity. While foot mass is typically difficult
adding or removing energy from the system. Terms that to reduce and drag couples the total mass to touchdown
will always be negative, thereby removing energy from the velocity for a given height, the force is extremely sensitive to
system, include, αf Dd , αT D∓ , αmech , αLO∓ , and αf Dr will the foot-ground interface stiffness which can be easily adjusted
be replaced with αi = −(1 − ηi ) indicating an effective with changes to the foot contact material; As an example,
efficiency measure (not necessarily a true efficiency). The assuming a touchdown velocity of 5.9 m/s (approximately 2
rebound control input must be αr∗ ≤ 1, as greater than meter drop) and a maximum foot compression of 0.3 mm,
one would indicate a average force greater than gravity, and the model predicts, a maximum foot acceleration of 1055 g’s
the robot would continue to rise indefinitely. Experimental (1128 N) at touchdown as compared to the maximum body
determination of these energies is discussed in appendix: acceleration during stance of 23 g’s (134 N); the touchdown
experimental energies. forces are over 8 times larger. Increasing the foot compression
Solving equation 12 for the rebound energy input ratio, αr∗ , to 0.9 mm reduces the foot acceleration to 319 g’s (338 N).
yields, This therefore decouples the foot mass from the body mass
allowing for independent scaling and design.
(αd∗ + ηF Dd )ηT D∓ (αs∗ + ηmech )ηLO∓
αr∗ = 2 − ηF Dr − ;
δrd
(13) III. ROBOT D ESIGN
The idealized simple model provides inspiration for a maxi-
where other jumping and hopping robots may choose to
mally efficient hopping robot. It has no joints or connections to
solve for the energy input ratio related to their input strat-
the elastic energy storage device to produce friction, assuming
egy. Considering all potential modifications to the LE, given
negligible foot mass, spring mass, drag, and spring material
aerodynamic based rebound energy input, increases in body
dissipation, the simple model would be 100% efficient. To
mass at δrd = 1 would result in increases to the effective
minimize the required energy input, and maximize efficiency
efficiencies ηF Dd,F Dr,T D∓,LO∓ . Therefore, from equation 13
and performance, the hopping robot design concept is centered
and assuming the mechanical losses, ηmech , scale with mass,
around the idealized model. The MultiMo-MHR (Table I), is
increasing body mass would result in decreasing the required
designed with a monopedal hopping leg for strong and efficient
rebound energy input ratio, αr∗ , and thus actuator size as a
hopping locomotion and a quadrotor configuration for energy
function of the body mass, allowing for high performance
input and control (Fig. 2a). The monopedal design produces a
hopping robot designs with sufficient payload for task specific
hopping leg with a single joint and not only minimal friction
components.
but with a significant decoupling of the friction from the
locomotion forces; as the locomotion forces are predominantly
B. Mass Scaling. parallel to the friction direction. Additionally, with the inherent
Scaling however, is usually assumed to apply to the entire simplicity of the design and primarily compressive forces on
robot; i.e., as body mass increases so too would the foot mass the structure, the design produces an extremely strong and
to withstand the increased locomotion forces exerted by the robust hopping leg; the leg, with a safety factor of two, can
body mass. This is a common misconception that stems from withstand the touchdown forces at terminal velocity (12.43
applying jumping robot concepts to hopping robots. Due to the m/s from 33.7 m drop) of the MultiMo-MHR.
touchdown in hopping robots, creating an additional spring- A major loss of energy in jumping and hopping robots is
mass system, the maximum force experienced by the foot is in the connection between the energy input device and the
typically not applied by the body during the stance phase and main elastic energy storage device. These mechanisms tend
is instead applied by the foot at touchdown (equations 2, 3). to be complex and experience high forces, which produces
Additionally, the natural frequency of the foot-ground interface significant resistance to the release of these forces. To remove
tends to be significantly greater than that of the body-foot. this impediment to hopping locomotion and maintain high ef-
 6

ficiency, the MultiMo-MHR is designed, as with the idealized B. Control.

model, with no connection between the main elastic energy Assuming the MulitMo-MHR operates as a quadrotor dur-
storage device and the energy input device, and therefore ing the aerial phase it can be modeled as such [57], [58].
these losses are entirely removed; while the MultiMo-MHR A simple linear controller is developed around the model
utilizes unconnected aerodynamic energy inputs, unconnected to control the system and input energy (Fig. 2e). Two on-
inertial energy inputs could be used as well. The energy stor- board IMUs (6-axis, 1000Hz, MPU6050, and 9-axis, 200 Hz,
age utilizes elastomers (rubber bands) for their considerably WT901) provide orientation information and a motion capture
higher specific energy (energy/mass), minimizing the spring system (Vicon, Vantage v5, 100Hz) provides the translational
mass (appendix: elastic specific energy). The MultiMo-MHR information and the thrust is provided by four quadrotor
utilized six rubber bands, which at max system stretch has motors (iFlight XING2 1404 4600 KV with 4030 props). The
an average specific energy of 23.87 J ∗ kg −1 for the entire horizontal positioning PID controllers are setup as cascades,
robot system; the robot is designed to carry sufficient rubber generating offset angles which are included in the roll and
bands to absorb the impact from terminal velocity (12.43 m/s pitch PID controllers whereas the yaw and the vertical position
at 33.7 m drop). Finally, the leg and foot designs are developed PID controllers operate alone. The control is separated into
to minimize the masses of these components, while being three components including stabilization (roll, pitch, and yaw),
sufficiently strong to withstand impact at terminal velocity. horizontal position (x-axis and y-axis), and LE input (z-
These collectively represent the foot of the simple model. axis), which can be individually activated. In this way LE
input (accumulation, dissipation, constant LE) is accomplished
through a constant feed-forward term included in the z-axis
A. Impact Mechanics. PID controller.
As the largest forces and torques experienced by the
MultiMo-MHR are during ground contact, a two degree-of- IV. R ESULTS AND D ISCUSSION
freedom model is developed to understand the behavior when To validate the energy accumulation concept in hopping
the robot contacts the ground at a slight angle to the hopping robots, experimental trials were conducted (n=805 hops).
axis. This will result in a torque applied to the robot body Figure 3a-e and supplementary video 1, show the MultiMo-
that may alter or hinder locomotion. Assuming the robot has MHR at varying drop heights and rebound control inputs spec-
impacted the ground, the high frequency impact dynamics ified as hd (αr∗ ) including unpowered, 0.5m(0%), 1.5m(0%),
have dissipated, and damping in the spring is negligible, the 2.5m(0%), 3.5m(0%), and powered, 0.5m(60%), 1m(55%),
robot can be modelled as a planar two degree-of-freedom 1m(60%), 1.5m(50%), 1.5m(55%), 1.5m(65%), 1.5m(85%)
system during the stance phase as, and 2m(55%) trials; 10 trials were conducted of each, each
with between 2-3 hops per trial for αr∗ = 85% and up to
k
r̈ = rθ̇2 − (r − r0 ) − g cos θ (14) 17 hops per trials at lower αr∗ (Fig. 3f). The experimental
mB
  procedure begins with the robot hovering at the desired drop
mB r mF rF mB
θ̈ = + g sin θ − 2 rθ̇ṙ (15) height. The drop is initiated once stabilized, and only the
IT IT IT stabilization controller is active. At contact, all controllers
are turned off for 60 msec - approximate time of ground
where, IT = IcmB + mB r2 + IcmF + mF rF2 is the total
contact - at which point the stabilization, horizontal position,
rotational inertia, r is the distance along the leg axis, r0 is
and constant energy input controllers are activated; the αr∗
the spring free length, IcmB is the rotational inertia of the
is a percent of the average duty cycle require to hover at
body about its center-of-mass (CM), IcmF is the rotational
the initiation of the drop. All controllers are active until the
inertia of the foot about its CM, and rF is the distance from
vertical velocity is reduced to less than 1 m/s, at which point
foot CM to the contact point. As shown in Fig. 2b-d, small
only the stabilization controller remains active up to the apex
angular deviations from vertical during ground contact can
of the rebound. The cycle then repeats for subsequent hops
lead to significant changes in the liftoff angle, required friction
without the initial hover and stabilization part. Figure 3g-i
coefficient, and vertical, horizontal, and rotational energies;
shows that the stabilization and horizontal position controllers,
e.g., a 6 m/s vertical velocity and 10 degree deviation from
in conjunction with the passive resistance of the foot to off-
vertical at ground contact would result in a liftoff angle of
axis ground contacts, are capable of maintaining the robot’s
45 degrees, required friction coefficient of 1, and vertical,
position over significant time periods and hopping cycles.
35%, horizontal, 44%, rotational, 4%, foot contact loss, 17%,
energies. This poses a significant challenge to increasing the
hopping height, and would also result from off-hopping-axis A. Energy Accumulation.
positioning of the center-of-mass. To minimize this effect, the Figure 3f shows that the MultiMo-MHR is able to ac-
foot is designed with four passive elastic rods to resist the cumulate energy at a critical rebound energy input ratio of
torque on the robot during ground contact; expanding the range αr∗ > 0.5 or half the force of gravity; where the up arrow
over which the robot is capable of operating. Additionally, indicates energy accumulation and the down arrow indicates
the added stability increases overall efficiency and facilitates energy dissipation. Assuming the robot begins in the gray
control as the controller need not overcome as significant a region of the plot, it will accumulate energy (increasing the
deviation from the desired behavior. rebound or subsequent drop height) until it reaches the δrd = 1
 7

a e f

1.25
Hopping Height Ratio, δrd=hr/hd

1.5
0.147 m Liftoff Vertical Velocity, (m/s)
5 Experimental Data
0.423 m 3

0.9
1.0
0.6

0.7
0.8

2.0
0.5
0.4
Down Up

Drop Height (m)

Jump 1 Jump 2 Jump 4 Jump 6 2
4
b

2.5
Jump 3 (3.941 m)
1 3

2
Critical αr*

3.282 m
1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
1.494 m Rebound Energy Input Ratio, αr*
1.058 m
1.688 m
g 1.6

z-Position (m)
1.2 Drop
0.8 Stance
Rebound
Jump 1 Jump 2 Jump 6 Jump 10 0.4
c Jump 1 Jump 2 0

d h 0.3
Position (m) 0.2
0.1
0
-0.1 x-Position
-0.2 y-Position

1.438 m 1.547 m i 8
0.221 m
Angle (deg)

4
0.737 m 0
-4
-8 Roll Pitch Yaw
0 5 10 15 20 25
Jump 1 Jump 2 Jump 6 Jump 10 Jump 1 Jump 2 Jump 3 Time (s)

Fig. 3. Experimental trials: a) Reduction in hopping height at an approximate 0.5 meter initial drop height and an energy input of 60% of the hover duty
cycle. b) Increase in hopping height at an approximate 1 meter initial drop height and an energy input of 60% of the hover duty cycle. c) Constant hopping
height at an approximate 1.5 meter initial drop height and an energy input of 55% of the hover duty cycle. d) Reduction in hopping height at an approximate
1.5 meter initial drop height and an energy input of 0% of the hover duty cycle. e) Increase in hopping height at an approximate 1.5 meter initial drop height
and an energy input of 85% of the hover duty cycle. f) Hopping height ratio of the experimental trials (n=805). g-i) Example trial with 17 hops showing the
6 degrees-of-freedom. Note: a-e) Snapshots of high-speed video with robots superimposed to show multiple phases. h-i) Symbols: x represents the hop apex,
circle represents the touchdown, and diamond represents the liftoff.

line, whereas, if it begins in the white region, it will dissipate hopping cycle efficiency, ηcyc , which can be extracted from
energy until it reaches the δrd = 1 line or a drop height of the αr∗ = 0 trials (Fig. 3f) and shows a range of between
zero; indicating cessation of hopping. 0.35 and 0.5; which is, as expected, drop height dependent.

B. Energy Dissipation.
C. Locomotion Energy Modification Characteristics.
While the paper has focused on energy accumulation, loco-
motion requires the robot to both accumulate and dissipate Extracting and analyzing the eight energy modifications to
energy. Through setting αr∗ greater than the critical value the LE from the experimental trials (appendix: experimen-
and within the white region of the plot (Fig. 3f) the robot tal energies, Fig. 4), highlights characteristics of each. The
is able to dissipate energy such that it will stabilize to a lower MultiMo-MHR’s fundamental design requires that energy be
hopping height. This feature allows for efficient control of the removed for stabilization during the drop phase. The results
hopping height as the robot can simply use its inefficiencies show the drop energy input ratio, αd∗ , reduces with increased
to remove energy over time as opposed to inputting energy to height indicating a relatively constant reduction of energy
change height. The maximum reduction in LE per cycle is the caused by stabilization. Therefore as hopping height increases,
 8

1.05 a 1.00
b e

Drag Efficiency (drop)

0.98 Function Fit
Experimental Data
Drop Efficiency

Experimental Data
1.00 0.99
0.94
3

Drag Efficency (rebound)

0.95 0.98
0.90

Drop height (m)

0.90 0.86 0.97
2
Mechanical Efficiency

0.96
c 0.96 d
0.85

Liftoff Efficiency
0.80 0.95
0.92
0.75 1 0.94
0.70
0.88
0.65 0.93
0.60 0.84
0 0.92
0 1 2 3 4 0 1 2 3 4 0 0.2 0.4 0.6 0.8
Drop Height (m) Drop Height (m) Energy Input Ratio

Fig. 4. Center-of-mass hopping energy modification. a) Assuming the control input at drop removes energy to stabilize, the control input can be seen as
an efficiency, αd∗ = 1 − ηd∗ , ηd∗ = 1 − Ed∗ /(mT ghd ). b) drag efficiency at drop, ηF Dd = (1 − EF Dd /(mT ghd )). c) Mechanical efficiency of
the hopping mechanisms, ηmech = 1 − Emech /(1/2mB vT 2 ). d) Liftoff efficiency due to losses in energy transfer between components, ηLO∓ =
D−
1 − ELO∓ /(1/2mB vLO 2 ). e) Drag efficiency at rebound given the drop height and rebound control energy input, ηF Dr = (1 − EF Dr /(mT ghr )).
−

and without perturbations, the energy reduction due to sta- will be the focus of future work. The experimental results
bilization will become negligible. The mechanical efficiency, also disprove the conclusion of a previous work, that hopping
ηmech , produces relatively constant values over the tested locomotion is only more efficient than flight at hopping heights
ranges suggesting both frictional losses and dissipation in the under 0.4 meters. Finally, of the five fundamental losses in
elastic material are proportional to the LE and less dependent hopping locomotion only drag shows a hopping height limiting
on the stretch length of the elastomer. The liftoff efficiency, behavior, suggesting that operation in environments without an
ηLO∓ , shows a decreasing value which may stabilize at atmosphere may allow continuous energy accumulation up to
high hopping heights suggesting, as expected, a coefficient the strength limits of the robot or the specific energy of the
of restitution type behavior for the impact between the body spring.
and foot. The ability to extract and analyze the individual While animals have shown hopping locomotion is an effi-
energies allows for future generations of the MultiMo-MHR to cient and effective means of traversing rugged terrain, robotic
maximize efficiency, energy accumulation, and performance. examples are limited. To advance the state-of-the-art, we
Moreover, the experimental loses suggest that only drag, which studied the fundamental design of hopping systems and the
shows constant decreasing efficiency with increasing drop hopping cycle energy. Our results demonstrate a new design
height, is responsible for the limit cycle behavior in which the paradigm for high-performance hopping robots. More broadly
hopping height stabilizes for a given input. As an example, this work creates a platform with sufficient payload (sensing,
assuming only touchdown losses, ηT D∓ which is constant, communication, computation) and performance to study artifi-
energy is accumulated continuously above the critical αr∗ ; cial intelligence, machine learning, and edge computing under
verified in simulation (Fig. S1). Therefore, in non-atmospheric highly dynamic locomotion conditions in a miniature robot.
environments energy can be accumulated indefinitely and is Additionally, as high-speed terrestrial locomotion dynamics
only limited by the maximum specific energy of the elastic converge to hopping type behaviors, the concepts developed
material and the strength of the robot. here, and those that could be developed with the platform, have
the potential to be broadly applicable to advancing terrestrial
V. S UMMARY robotics in general.
This work has redeveloped hopping robot design and energy
accumulation principles from an idealized model to maximize A PPENDIX A
efficiency, robustness, energy input, and performance. The E XPERIMENTAL E NERGIES
design concepts that developed the MultiMo-MHR include,
Determining the αi ’s and ηi ’s experimentally is achieved
efficient design (limited joints, unconnected energy storage
by first determining the drag efficiencies from the trajectories.
and energy input systems), robust design (linear leg, limited
The effective drag efficiencies of drop, ηF Dd and rebound,
joints, axial forces, mass scaling), energy accumulative design
ηF Dr are determined as,
(aerodynamic forces, energy input at rebound), and controls
R hd
design (management of ground contact mechanics, energy FD dh
input). The MultiMo-MHR is able to hop over 6 times greater ηF Dd = 1 − 0 (16)
mT ghd
than the previous state-of-the-art to heights of nearly 4 meters. R hr
This performance is not limited by the energy input, robust- FD dh
ηF Dr = 1 − 0 . (17)
ness, or efficiency but by the ground contact mechanics which mT ghr
 9

αr* : 99% 90% 80% 70% 60% 50% 40% 12 Trial 1 Trial 2 Trial 3 Mean

Stress (MPa)
70 8
Model
60
mB 4
Body Height (m)

50
0
0 1 2 3 4 5 6 7
40 Strain
30 mF Fig. S2. Rubber band stress vs. strain plot.
20
10
0 [2] C. Zhang, W. Zou, L. Ma, and Z. Wang, “Biologically inspired jumping
0 5 Time (s) 10 15 20 robots: A comprehensive review,” Robotics and Autonomous Systems,
vol. 124, p. 103362, 2020.
[3] E. W. Hawkes, C. Xiao, R.-a. Peloquin, C. Keeley, M. R. Begley, M. T.
Fig. S1. Simulation results of the MultiMo-MHR with only touchdown losses. Pope, and G. Niemeyer, “Engineered jumpers overcome biological limits
via work multiplication,” Nature, vol. 604, no. April, 2022.
[4] Boston Dynamics, “Sandflea,” 2013.
The drag force is defined as, [5] M. Kovac, M. Fuchs, A. Guignard, J.-C. Zufferey, and D. Floreano,
“A miniature 7g jumping robot,” IEEE International Conference on
Robotics and Automation, pp. 373–378, may 2008.
FD = 1/2CD Aρv 2 (18) [6] V. Zaitsev, O. Gvirsman, U. Ben Hanan, A. Weiss, A. Ayali, and
G. Kosa, “A locust-inspired miniature jumping robot,” Bioinspiration
where, the drag coefficient and effective area, CD A = & Biomimetics, vol. 10, no. 6, p. 066012, 2015.
−0.001893 ∗ v + 0.072122, are determined experimentally [7] B. Shin, H.-Y. Kim, and K.-J. Cho, “Towards a biologically inspired
and fit to the absolute value of the system velocity, v, and small-scale water jumping robot,” IEEE RAS & EMBS International
Conference on Biomedical Robotics and Biomechatronics, pp. 127–131,
total mass. This assumes the velocity vector is approximately oct 2008.
co-linear with the z-axis of the robot and v ≤ 7 m/s. The [8] W. a. Churaman, A. P. Gerratt, and S. Bergbreiter, “First leaps
velocity and air density, ρ, then determine the drag force. toward jumping microrobots,” IEEE/RSJ International Conference on
Intelligent Robots and Systems, vol. 2, no. 2, pp. 1680–1686, sep 2011.
Subtracting the drag efficiencies from the total energy change [9] M. Noh, S.-W. Kim, S. An, J.-S. Koh, and K.-J. Cho, “Flea-Inspired
in the aerial phases (drop and rebound) can yield the control Catapult Mechanism for Miniature Jumping Robots,” IEEE Transactions
inputs (αd∗ , αr∗ ) in each phase. This does however, couple on Robotics, vol. 28, no. 5, pp. 1007–1018, oct 2012.
[10] W. a. Churaman, L. J. Currano, C. J. Morris, J. E. Rajkowski,
the drag losses to the control inputs. As the ground contact and S. Bergbreiter, “The First Launch of an Autonomous Thrust-
time tends to be consistent across different hopping heights in Driven Microrobot Using Nanoporous Energetic Silicon,” Journal of
a specific robot, due to its relation to the natural frequency, the Microelectromechanical Systems, vol. 21, no. 1, pp. 198–205, feb 2012.
[11] J. S. Koh, S. P. Jung, M. Noh, S. W. Kim, and K. J. Cho, “Flea inspired
stance control input, αs∗ can be determined. The mechanical catapult mechanism with active energy storage and release for small
efficiency, ηmech , can then be determined by subtracting the scale jumping robot,” IEEE International Conference on Robotics and
stance control input, αs∗ . The touchdown efficiency, ηT D∓ , Automation, pp. 26–31, 2013.
[12] F. Li, G. Bonsignori, U. Scarfogliero, D. Chen, C. Stefanini, W. Lui,
is calculated whereas the liftoff efficiency, ηLO∓ , can be P. Dario, and X. Fu, “Jumping mini-robot with bio-inspired legs,” IEEE
determine from trials with zero rebound control input, αr∗ . International Conference on Robotics and Biomimetics, pp. 933–938,
Using the equations from appendix: robot energy and the feb 2009.
[13] J. Zhao, J. Xu, B. Gao, N. Xi, F. J. Cintr, M. W. Mutka, and L. Xiao,
motion capture data from the experimental trials, the individual “MSU Jumper: A Single-Motor-Actuated Miniature Steerable Jumping
energies are shown in Fig. 4. Robot,” IEEE Transactions on Robotics, vol. 29, no. 3, pp. 602–614,
2013.
[14] M. M. Plecnik, D. W. Haldane, J. K. Yim, and R. S. Fearing, “Design
A PPENDIX B exploration and kinematic tuning of a power modulating jumping
E LASTIC S PECIFIC E NERGY monopod,” Journal of Mechanisms and Robotics, vol. 9, no. 1, pp. 1–13,
2017.
Experimental testing of the elastomers, Fig. S2, shows the [15] P. Fiorini and J. Burdick, “The Development of Hopping Capabilities
for Small Robots,” Autonomous Robots, pp. 239–254, 2003.
maximum specific energy is 15800.3 J/kg. The utilized bands [16] J. Zhao, R. Yang, N. Xi, B. Gao, X. Fan, M. W. Mutka, and
were found to have an average max stress and strain at 11.2 L. Xiao, “Development of a miniature self-stabilization jumping robot,”
MPa and 6.2 respectively. For our robot platform the bands can IEEE/RSJ International Conference on Intelligent Robots and Systems,
pp. 2217–2222, oct 2009.
only see a maximum strain of 2.1. At that point an average [17] U. Scarfogliero, C. Stefanini, and P. Dario, “Design and Development
stress of 1.2 MPa is expected and an average specific energy of the Long-Jumping ” Grillo ” Mini Robot,” IEEE International
of 1307.9 J/kg. The robot platform possesses 6 bands which Conference on Robotics and Automation, no. April, pp. 10–14, 2007.
[18] T. Ho and S. Lee, “A Novel Design of a Robot That Can Jump and
can store at the strain of 2.1 an average energy of 16.63 J, at Roll with a Single Actuator,” IEEE/RSJ International Conference on
the total robot mass of 696.55 g that correlates to an average Intelligent Robots and Systems, pp. 908–913, 2012.
total system specific energy of 23.87 J/kg. [19] J. Aguilar and D. I. Goldman, “Robophysical study of jumping dynamics
on granular media,” Nature Physics, vol. 12, no. 3, pp. 278–283, 2016.
[20] A. Yamada, M. Wataril, H. Mochiyama, and H. Fujimoto, “A Jumping
R EFERENCES Robot based on the Closed Elastica,” International Symposium on Micro-
NanoMechatronics and Human Science, pp. 604–609, 2007.
[1] R. Armour, K. Paskins, A. Bowyer, J. Vincent, W. Megill, and [21] A. Yamada, M. Watari, H. Mochiyama, and H. Fujimoto, “An
R. Bomphrey, “Jumping robots: a biomimetic solution to locomotion asymmetric robotic catapult based on the closed elastica for jumping
across rough terrain.” Bioinspiration & biomimetics, vol. 2, no. 3, pp. robot,” IEEE International Conference on Robotics and Automation,
65–82, sep 2007. pp. 232–237, may 2008.
 10

[22] A. Yamada, H. Mameda, H. Mochiyama, and H. Fujimoto, “A Compact [45] N. W. Bartlett, M. T. Tolley, J. T. B. Overvelde, J. C. Weaver,
Jumping Robot utilizing Snap-through Buckling with Bend and Twist,” B. Mosadegh, K. Bertoldi, G. M. Whitesides, and R. J. Wood, “A 3D-
IEEE/RSJ International Conference on Intelligent Robots and Systems, printed, functionally graded soft robot powered by combustion,” Science,
pp. 389–394, 2010. vol. 349, no. July, pp. 161–165, 2015.
[23] M. Kovac, M. Schlegel, J.-C. Zufferey, and D. Floreano, “A miniature [46] M. H. Raibert, “Hopping in Legged Systems—Modeling and Simulation
jumping robot with self-recovery capabilities,” IEEE/RSJ International for the Two-Dimensional One-Legged Case,” IEEE Transactions on
Conference on Intelligent Robots and Systems, pp. 583–588, oct 2009. Systems, Man and Cybernetics, vol. SMC-14, no. 3, pp. 451–463, 1984.
[24] M. Kovac, M. Schlegel, J.-C. Zufferey, and D. Floreano, “Steerable [47] M. H. Raibert, H. B. Brown, and M. Chepponis, “Experiments in
miniature jumping robot,” Autonomous Robots, vol. 28, no. 3, pp. Balance with a 3D One-Legged Hopping Machine,” The International
295–306, dec 2009. Journal of Robotics Research, vol. 3, no. 2, pp. 75–92, 1984.
[25] J. Zhao, N. Xi, B. Gao, M. W. Mutka, and L. Xiao, “Design and testing [48] D. W. Haldane, M. M. Plecnik, J. K. Yim, and R. S. Fearing, “Robotic
of a controllable miniature jumping robot,” IEEE/RSJ International vertical jumping agility via Series-Elastic power modulation,” Science
Conference on Intelligent Robots and Systems, pp. 3346–3351, oct Robotics, vol. 1, no. 1, 2016.
2010. [49] E. Ambrose and A. D. Ames, “Improved Performance on Moving-
[26] P. Fiorini, S. Hayati, M. Heverly, and J. Gensler, “A hopping robot for Mass Hopping Robots with Parallel Elasticity,” Proceedings - IEEE
planetary exploration,” IEEE Aerospace Conference Proceedings, pp. International Conference on Robotics and Automation, pp. 2457–2463,
153–158 vol.2, 1999. 2020.
[27] J. Zhao, N. Xi, B. Gao, M. W. Mutka, and L. Xiao, “Development [50] D. W. Haldane, J. K. Yim, and R. S. Fearing, “Repetitive extreme-
of a controllable and continuous jumping robot,” IEEE International acceleration ( 14-g ) spatial jumping with Salto-1P,” Submitted to IEEE
Conference on Robotics and Automation, pp. 4614–4619, may 2011. Int. Conf. Intell. Robots. Syst., pp. 3345–3351, 2017.
[28] J. Zhao, T. Zhao, N. Xi, M. W. Mutka, and L. Xiao, “MSU Tailbot [51] J. K. Yim, B. R. P. Singh, E. K. Wang, R. Featherstone, and R. S.
: Controlling Aerial Maneuver of a Miniature-Tailed Jumping Robot,” Fearing, “Precision robotic leaping and landing using stance-phase
IEEE/ASME Transactions on Mechatronics, pp. 1–12, 2015. balance,” IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 3422–
[29] M. A. Woodward and M. Sitti, “Morphological intelligence counters 3429, 2020.
foot slipping in the desert locust and dynamic robots,” Proceedings of [52] A. Sayyad, B. Seth, and P. Seshu, “Single-legged hopping robotics
the National Academy of Sciences, p. 201804239, 2018. research - A review,” Robotica, vol. 25, no. 5, pp. 587–613, 2007.
[30] M. A. Woodward and M. Sitti, “Design of a miniature integrated [53] B. A. Christensen, D. C. Lin, M. J. Schwaner, and C. P. McGowan,
multi-modal jumping and gliding robot,” IEEE/RSJ International “Elastic energy storage across speeds during steady-state hopping of
Conference on Intelligent Robots and Systems, pp. 556–561, sep 2011. desert kangaroo rats (Dipodomys deserti),” Journal of Experimental
[31] M. Kovac, O. Fauria, J.-C. Zufferey, and D. Floreano, “The EPFL Biology, vol. 225, no. 2, 2022.
jumpglider: A hybrid jumping and gliding robot with rigid or folding [54] H. Okubo and E. Nakano, “Design of a Jumping Machine Using Self-
wings,” IEEE International Conference on Robotics and Biomimetics, energizing Spring,” pp. 186–191, 1996.
pp. 1503–1508, dec 2011. [55] U. Scarfogliero, C. Stefanini, and P. Dario, “A bioinspired concept for
[32] M. A. Woodward and M. Sitti, “MultiMo-Bat: A biologically inspired high efficiency locomotion in micro robots: the jumping Robot Grillo,”
integrated jumping-gliding robot,” The International Journal of Robotics IEEE International Conference on Robotics and Automation, no. May,
Research, vol. 33, no. September, pp. 1511–1529, 2014. pp. 4037–4042, 2006.
[33] A. L. Desbiens, M. Pope, F. Berg, Z. E. Teoh, J. Lee, and M. Cutkosky, [56] R. M. Alexander, Principles of Animal Locomotion. Princeton Univer-
“Efficient Jumpgliding : Theory and Design Considerations,” IEEE sity Press, 2003.
International Conference on Robotics and Automation, pp. 4436–4443, [57] S. Bouabdallah, R. Siegwart, and G. Caprari, “Design and control of an
2013. indoor coaxial helicopter,” IEEE International Conference on Intelligent
[34] A. L. Desbiens, M. T. Pope, D. L. Christensen, E. W. Hawkes, and Robots and Systems, no. April, pp. 2930–2935, 2006.
M. R. Cutkosky, “Design principles for efficient, repeated jumpgliding.” [58] R. Mahony, V. Kumar, and P. Corke, “Multirotor aerial vehicles:
Bioinspiration & biomimetics, vol. 9, no. 2, p. 025009, 2014. Modeling, estimation, and control of quadrotor,” IEEE Robotics and
[35] A. Beck, V. Zaitsev, U. B. Hanan, G. Kosa, A. Ayali, and A. Weiss, Automation Magazine, vol. 19, no. 3, pp. 20–32, 2012.
“Jump stabilization and landing control by wing-spreading of a locust-
inspired jumper,” Bioinspiration & Biomimetics, vol. 12, 2017.
[36] M. A. Woodward and M. Sitti, “Tailored Magnetic Springs for Shape-
Memory Alloy Actuated Mechanisms in Miniature Robots,” IEEE
Transactions on Robotics, vol. 35, no. 3, pp. 589–601, 2019.
[37] B. Zhu, J. Xu, A. Charway, and D. Saldana, “PogoDrone: Design,
Model, and Control of a Jumping Quadrotor,” Proceedings - IEEE
International Conference on Robotics and Automation, pp. 2031–2037,
2022.
[38] B. G. A. Lambrecht, A. D. Horchler, and R. D. Quinn, “A Small ,
Insect-Inspired Robot that Runs and Jumps,” IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp. 40–45, 2005.
[39] A. M. Johnson and D. E. Koditschek, “Toward a vocabulary of legged
leaping,” Proceedings - IEEE International Conference on Robotics and
Automation, pp. 2568–2575, 2013.
[40] A. L. Brill, A. De, A. M. Johnson, and D. E. Koditschek, “Tail-assisted
rigid and compliant legged leaping,” IEEE International Conference on
Intelligent Robots and Systems, vol. 2015-Decem, pp. 6304–6311, 2015.
[41] G. P. Jung, C. S. Casarez, S. P. Jung, R. S. Fearing, and K. J. Cho,
“An integrated jumping-crawling robot using height-adjustable jumping
module,” Proceedings - IEEE International Conference on Robotics and
Automation, vol. 2016-June, pp. 4680–4685, 2016.
[42] S. a. Stoeter and N. Papanikolopoulos, “Autonomous stair-climbing
with miniature jumping robots.” IEEE Transactions on Systems, Man,
and Cybernetics, vol. 35, no. 2, pp. 313–325, apr 2005.
[43] Y. Ho, S. Kyoung, T. Jin, J. H. Lee, J. Y. Choi, C. H. Yim, and D. H.
Kim, “Mecianism and Control of a Jumping Robot,” in International
Conference on Control, Automation and Systems 2007, 2007, pp. 2499–
2502.
[44] M. T. Tolley, R. F. Shepherd, M. Karpelson, N. W. Bartlett, K. C.
Galloway, M. Wehner, R. Nunes, G. M. Whitesides, and R. J. Wood,
“An Untethered Jumping Soft Robot *,” IEEE International Conference
on Intelligent Robots and Systems, pp. 4–9, 2014.