You are on page 1of 6

2016 American Control Conference (ACC)

Boston Marriott Copley Place


July 6-8, 2016. Boston, MA, USA

Design and Control of an Electrically Assisted Kick Scooter


Matteo Corno, Fabio Busnelli, Sergio M. Savaresi

Abstract— Traffic congestion, energy efficiency and environmen- Electric Human-Powered Hybrid (see for example [4], [6]–
tal issues are fueling the interest in Light Electric Vehicles [8], [11]–[13], [15]). EPAC’s are a very effective solution for
(LEV’s). Electric Human-Powered Hybrid vehicles show a great short range commute, but they are not well-suited for mixed-
potential: they are cost effective, safe, easy to use and have a
small footprint. In this paper, we discuss an electrically assisted transportation modalities: for example, one could use the
kick scooter. Electric scooters are common on the market train system to get to the city and may still need a convenient
but, being throttle-controlled, they sit in a legislative gray vehicle to cover the last mile. EPAC’s are bulky and heavy
area between human powered vehicles and electric vehicles. and cannot easily be carried on a train during rush hours.
The proposed design solves this issue by designing a genuine
electrically assisted scooter. The proposed design does not have
any human machine interface but controls the electric assistance This paper addresses this issue presenting an electrically
in a transparent way. The user only needs to kick the scooter
as she would do with a normal scooter and the control system assisted vehicle suitable as a last mile means of transporta-
smoothly delivers the assistance. The proposed solution is based tion. In order to achieve this objective the vehicle needs to
on a kick detection algorithm and a closed-loop control of the achieve two objectives 1) it needs to be less bulky than a
vehicle velocity. The details of the control system are discussed bicycle 2) it needs be an electrically assisted vehicle and
in the paper and an extensive experimental validation shows not an electrically powered vehicle. The kick scooter is an
that the electric assistance cuts in half the effort necessary to
operate the vehicle with a battery consumption of 4 Wh/km interesting candidate: it is light, small and foldable. Many
(half that of an electrically assisted bicycle). electric kick scooters are already available on the market.
The available kick scooters do not however fulfill the second
I. I NTRODUCTION AND M OTIVATION requirement ( [2], [10]). In these vehicles, the electric motor
is controlled by a throttle, thus making them legally electric
The need for reducing pollution and improving city mobility vehicles. In many countries, this subtle difference is being
is propelling the research into alternative mobility modalities. overlooked by traffic officers, but it is expected that, as
Car powertrain electrification (or hybridization) is a common these vehicles gain popularity, the rules will be more strictly
approach that has received attention. It has been shown to applied.
be viable for medium-to-long driving range: electric vehicles
improve efficiency and reduce emissions. However electric The proposed design avoids the risk of being mistakenly
cars do not address the problem of mobility and congestion classified as an electric vehicle avoiding the use of a throttle.
especially critical in European cities. Light transports that The control system detects each kick and assists the rider
also have a small footprint are necessary. As a matter of by extending each kick’s reach. The proposed system does
fact, small electric vehicles are becoming more and more not have any external human-machine interface: it does not
common for short driving ranges. Small electric vehicles have a throttle, nor activation buttons nor switches. This
still have one main drawback: in most countries they require guarantees that from the rider point of view the scooter is
some sort of driving permit, registration, insurance and a operated and feels like a traditional scooter. The net effect
helmet. These small nuisances, required to guarantee safety, is however that of reducing the rider’s fatigue and effort.
are disincentivizing the use of small electric vehicles. The From the control point of view the challenges involved
price for a street legal electric scooter could be as low as in designing a control system for this kind of vehicle are
2500 euros, but one could then need to account for insurance superior than those involved in EPAC’s. Matter-of-factly, in
costs up to 400 euros/years in some cities. This limitation EPAC’s, human power is applied in a continuous and smooth
is addressed by Electric Human-Powered Hybrid vehicles. fashion; in the kick scooter the human power is impulsive.
This kind of vehicles does not require a registration (under This calls for a completely different control architecture.
most legislations) as they are powered by the human in a
very intuitive way: the electric motor assist the user and help
reduce the physical exertion. In the past few years, Electri- The paper is structured as follows: in Section II, the exper-
cally Power Assisted Bicycles (EPAC’s) are experiencing a imental set-up is described and modeled. In Section III, the
considerable commercial success [16]. They cost less than challenges of designing a control system for an impulsive
an electric scooter and do not have any mandatory insurance vehicle are discussed and addressed. In Section IV, the
cost. This success has also fueled a deal of research into vehicle is validated on a real last mile use profile and the
reduction in physical effort is shown. Section V concludes
Corno, Busnelli and Savaresi are with the Dipartimento di Elettronica,
Informazione e Bioingegneria, Politecnico di Milano, Piazza Leonardo da the paper. Part of the methods described in this paper are
Vinci 32, 20133 Milano, Italy. matteo.corno@polimi.it object of a patent [5].

978-1-4673-8682-1/$31.00 ©2016 AACC 5290


II. E XPERIMENTAL SET- UP AND M ODELING where the vehicle is assumed to be driving on a flat surface.
M is the total vehicle + rider mass, v is the vehicle velocity,
Figure 1 represents a prototype of the vehicle under study. A (α, β, γ) are the parameters describing the road, gear and
aerodynamic friction, Ke is the motor current constant, r
the wheel radius and Fbrake is the force applied through
the friction brakes. The model depends on a number of
parameters and an unknown input (the braking force). Some
of the model parameters are directly measured: M r and Ke .
The friction parameters are identified using a coasting down
experiment: the scooter is manually accelerated to a velocity
of 15 km/h and let slow down without any braking from
the brakes nor the motor. Figure 2 shows the results of the
coasting down experiment. Given the relatively low velocities

-0.05
measurement
linear fitting
-0.1 quadratic fitting

acceleration [m/s ]
2
-0.15

-0.2

-0.25
Fig. 1. The assisted kick scooter prototype.
-0.3
commercially foldable kick scooter frame has been employed 0 1 2 3 4 5
speed [m/s]
and equipped with an all-in-one EPAC kit. The scooter is
equipped with: Fig. 2. Scooter coasting down experiment.

• A brushless motor. The motor delivers a nominal power that the vehicle reaches, the quadratic term due to the aero-
of 250 W. The motor is rigidly fixed to the wheel, dynamic drag is dominated by the viscous friction (bearing
so that it always spins at the velocity of the wheel friction) and the static friction (tyre rolling friction). Figure
(and consequently the velocity of the motor is statically 3 plots the validation of the model when the motor current
related to the velocity of the vehicle). tracks a triangular wave. The simulated velocity matches the
• A custom made Electric Control Unit (ECU). The ECU
implements the high level control algorithm and the
5
low level motor drive control. It also has bluetooth
motor current [A]

4
connectivity.
3
• Battery Pack. A relatively small 30 V (160 Wh) Li-ion
2
battery pack. The battery pack is equipped with a Bat-
tery Management System (BMS); which, among other 1

safety related checks (single cell voltage, temperature 0

and over-current), runs a Kalman-based SoC estimation 45 50 55 60 65 70 75


5
algorithm [14]. measured
• A wireless brake switch. The switch sends a signal to 4.5 simulated
speed [m/s]

the ECU whenever the brake is pulled. This enables 4


to cut assistance off every time the brake is pulled (as
3.5
required by the relevant legislation)
• A Mio Global wrist Heart Rate Monitor. This sensor is 3
connected to a smartphone. It is employed for validation
2.5
and analysis purposes and not for control. 45 50 55 60 65 70 75
time [s]
In the context of this work, the main objective is that of
providing assistance along the longitudinal direction; thus Fig. 3. Scooter coasting down experiment.
the vehicle dynamics can be modeled as:
Ke measured one with a satisfactory accuracy. The identification
M v̇ = −αv 2 − βv − γ + i − Fbrake (1) and validation experiments have been performed on a flat
r
5291
asphalt road; naturally it has been found that the results will that no external force (nor tractive nor braking) is applied
vary if the scooter is ridden on a different surface and - to after the velocity has reached its peak. If a passive scooter
a lesser extend - by a different rider. The control presented feeling has to be preserved, the traction should be activated
in the what follows will account for this variability so to only at the end of the sequence, not right after the velocity
guarantee an adequate level of robustness. peak. If motor was to be turned on right after the peak, the
vehicle would accelerate more than a passive scooter and
III. K ICK S COOTER C ONTROL would loose its purely assistive nature.

The main objective of the controller is that of operating the In order to provide the correct feeling, the kick event is
motor so that it extends the reach of each single push. The divided into several different phases:
rationale is that the motor should never cause an acceleration
greater than what the rider could achieve on a traditional • pre-kick it precedes the sharp increase of velocity. The
scooter; the motor can only reduce the deceleration. This rider is extending her leg but the foot has not come
guarantees a greater safety. The task is made difficult by into contact with the ground. The velocity increases as
two issues: 1) the controller cannot rely on any direct a consequence of conservation of momentum.
input from the rider. 2) The way in which the human • kick the leg is exerting a force against the ground and
traction power is transmitted to the ground is impulsive the velocity sharply increases. Starting from this phase
by nature, thus the control algorithms devised for assisted the motor needs to be turned off, to avoid unwanted
bicycles cannot be directly applied to this case. In order to accelerations.
achieve this objective, the controller is organized in several • release the leg is retracted, the quick movement of the
modules as shown in Figure 4. In the schematics, the vehicle leg slows down the vehicle.
• rebound the leg’s movement has now come to a stop
and is almost parallel to the other leg. As the leg slows
down, momentum is transferred to the vehicle which
accelerates.
• coasting-down the kick event is now complete and the
rider is fairly stable on the scooter, the scooter slows
down due to friction. The motor is operated to extend
Fig. 4. Scooter coasting down experiment. this phase.

block represents the scooter dynamics. The ax estimation


The above sequence of phases is detected using a finite state
block simply estimates the vehicle longitudinal acceleration
machine represented in Figure 6. As it can be seen most
through numerical differentiation of the wheel velocity. R is a
transitions are based on acceleration thresholds. Also note
closed-loop velocity controller; the velocity reference block
that the brake switch has the authority to turn off assistance at
generates a suitable reference velocity. The kick detection
any time. The transition from pre-kick to kick deservers more
block determines whether to activate the velocity control
comments as it is the fundamental transition that activates a
and thus the assistance. In the following the most important
cascade of important events. The transition is based on ∆v
subcomponents are detailed out.

A. Kick Detection
14
As the name implies, the goal of the kick detection module is
that of determining when the rider is pushing the vehicle. The 13
task is made difficult by the characteristic of the propulsion. 12
In a normal kick event, the rider imparts the forward force
by extending her leg backward. When the rider extends the 11
speed [km/h]

leg, the foot makes contact with the ground and the force is
10
transmitted to the vehicle. After this phase, the leg is usually
coasting down
coasting down

retracted to be parallel to the leg that supports the rider’s 9


weight. This, since the mass of the rider is considerably
rebound
pre-kick

release

8
higher than that of the scooter, slows down the vehicle (total
kick

momentum conservation). This phenomenon is depicted in 7


Figure 5. The figure shows a kick without any additional
assistance from the motor. The figure clearly shows that when 6
37.5 38 38.5 39 39.5 40
the vehicle is pushed the velocity sharply increases. After the time [s]
initial phase, the velocity drops very rapidly to then increase
again due to the movement of the rider’s leg and body. Note Fig. 5. Analysis of a single push event.

5292
computed as: Z t
12
∆v(t) = ax (t)dt

speed [km/h]
(2)
tstart
10
where tstart is the time instant at which the machine tran-
sitioned into the pre-kick phase. Instant by instant ∆v is reference
8 simulated
compared against a velocity dependent threshold. If ∆v is measured
greater than the threshold then the kick is confirmed. The 6
threshold is velocity dependent; in particular the threshold 150 2 4 6 8 10

decreases as the velocity increases; in this way the robustness

motor current [A]


of the algorithm is increased without loosing the capability of 10
detecting kicks at high speed. Matter-of-factly, the velocity
is more easily perturbed by disturbances at low speed than
at high speed. If a constant threshold were to be used, 5
this would either cause false or miss detections. Note that
∆v could also be computed as the difference between the 0
current velocity and the velocity at which the last transition 0 2 4 6 8 10
happened. The transition from kick to coasting down, which time [s]
happens if the rider does not retract the leg, is based on a Fig. 7. Closed-loop validation of the velocity controller.
similar condition.

B. Velocity Controller is moving. In order to provide a bump-less activation of the


controller, it is initialized at the steady state current required
The velocity controller is a standard closed-loop controller to move at the initial reference velocity of activation.
tuned on the identified dynamics. The controller is tuned
to achieve a closed-loop bandwidth of 0.2 Hz and a phase
margin of 30◦ . Sensitivity analysis shows that the controller C. Velocity Reference
is stable for a total mass ranging from 40 to 120 kg and a
The velocity reference generation simply generates a desired
tenfold variation of the viscous friction parameter identified
behavior for the coasting down. Different philosophies may
in Section II. Figure 7 plots the comparison between the
be applied to generare the reference velocity. Through trial
simulated and experimental results for a velocity reference
and error and users’ feedback, it was found that the most
step: the reference is accurately tracked as expected from
accepted reference is a linearly decreasing reference. In
design. It is interesting to note the behavior immediately after
particular, the slope of the reference is computed at the motor
the change of reference. The real vehicle exhibits an initial
activation time instant so that regardless of the initial velocity
behavior that is determined by the movement of the rider’s
the duration in time of each kick is the same. Figure 8 plots
body (who responds to the jerk). The model presented in II
the validation of the entire system after a single kick event.
does not account for this; the velocity controller is however
As it can be seen, the kick is correctly detected and the
robust to this phenomenon.
velocity tracked accurately without negatively affecting the
According to the operation principle explained above, the natural behavior of a kick.
velocity controller needs to be turned on when the vehicle
IV. VALIDATION

This section validates the proposed vehicle on a realistic


mission profile. In particular the data are analyzed from

16
velocity
15 reference
velocity [km/h]

14

13

12

11

10
52.5 53 53.5 54 54.5 55 55.5 56
time [s]

Fig. 6. Finite State Machine determining the kick event phases. Fig. 8. Closed-loop validation of the velocity controller.

5293
20 required kicks is considerably reduced, from an average
level of 250 with no assistance to 122 with a 1%
average trip speed [km/h]

15 assistance. This intuitively leads to a more pleasurable


riding experience.
• As the assistance level is increased the number of
10
required kicks decreases following a diminishing return
trend. This is explained by the fact the test is performed
5 on a real road. If the coasting down is extended beyond
a certain threshold, the advantage of that increase is lost
to the fact that the rider needs to slow down in order to
0
013 6 40 70 100 obey the traffic rules.
assistance level [%]
Another important aspect is the average energy consumption
Fig. 9. Trip average velocity for different assistance levels. of the vehicle. The average energy per km is represented in
Figure 11 as a function of the assistance level. A diminishing

the battery energy and rider’s effort points of view. A 2.5 8


km/h urban trip has been defined. The experiments have

battery consumption [Wh/km]


been executed during the day on open public roads, thus
6
they account for traffic. The trip has been executed with
different settings: passive scooter (same set-up but without
any assistance) and 6 different levels of assistance indicated 4
as percentage. The assistance level is normalized between
a maximum duration of the coasting down of 1100 s and 2
a minimum of 45 s. For each setting 10 runs have been
repeated in order to account for the variability introduced by
0
the traffic. For each test all the relevant electric and kinematic 013 6 40 70 100
measures have been logged along with the riders’ heart rate. assistance level [%]
The heart rate was logged also during a 10 minute rest period
Fig. 11. Average energy consumption for different assistance levels.
after each run.
The riders were instructed to try and keep the same average return trend is observed also in this case. Note the 0%
velocity with the different settings. Figure 9 plots the results level of assistance still consumes around 0.05 Wh/km for
in terms of velocity. Although it is subject to a certain level powering logging and communication devices. In order to
of variability, the average velocities are comparable. Note better appreciate the absolute values shown in figure consider
that the one with the lowest velocity is the case without that an EPAC’s can reach a battery energy consumption as
assistance. In this case the riders could not keep up with high as 8.5 Wh/km [1]; the assisted scooter tops at 6.5
the target velocity. As the average velocities are similar, it is Wh/km. Given the battery pack employed in the vehicle,
interesting to consider the number of kicks required to cover the vehicle has a range from 25 to 40 km, depending on the
the 2.5 km/h trip. These are show in Figure 10. From the level of assistance.
The analysis is concluded by considering a measure of rider’s
300
exertion or fatigue. Measuring fatigue is not an easy task [3];
250 however heart rate can be employed as a useful proxy for
fatigue [9]. In order to evaluate riders’ fatigue, the following
number of kicks []

200 cost function is thus proposed:


(
150 0 HR(t) < HR ˜
JHR = Rt
˜ 2 ˜
100 0 q · (HR(t) − HR) HR(t) >= HR
(3)
50
where HR ˜ is a baseline heart rate under which fatigue is
0 not accumulated. In this work, it is set at 100 BPM. q is
013 6 40 70 100 a weight index that controls the penalty to high heart rate
assistance level [%]
vs low heart rate. Here q is set to 1/802. The heart rate is
Fig. 10. Number of requires kicks for different assistance levels. known to have a slow dynamics [6], this means that even
after the end of the physical exertion, the subject needs a
figure the following remarks are due: rest period before her baseline is reached. This rest period
needs to be accounted in the evaluation of (3) in order to have
• Even for a small level of assistance, the number of comparable results. Figure 12 plots the relationship between

5294
JHR and the average battery consumption The following VI. ACKNOWLEDGMENTS
final remarks can be drawn from figure (1) The assistance
is very effective in reducing physical exertion. Even the The authors would like to thank ZEHUS srl. sole owner of
lowest level of assistance reduces JHR to almost half. (2) the patent [5] in the persons of P. Lisanti and D. Berretta for
JHR exhibits a linear trend with the average battery energy the technical support and to S. Gelmini and A. Mosconi for
consumption. There is a roughly linear exchange between managing the human-in-the-loop experiments.
human exertion and battery energy.
R EFERENCES
V. C ONCLUSIONS
[1] G. Alli, S. Formentin, and S. Savaresi. On the suitability of EPAC’s
in urban use. In Mechatronic Systems, pages 277–284, 2010.
In the present paper an electrically assisted kick scooter is
presented. In particular, the focus is on the design of the [2] A Brankovic, D Berretta, S Formentin, and S.M Corno, M Savaresi.
Modeling and speed limitation control of an electric kick scooterr.
control system. The control system is designed to provide In Proceeding of the European Control Conference, pages 764–769,
an extremely intuitive user experience. The rider does not 2015.
need to operate throttles or switches, she only needs to kick [3] George Austin Brooks, Thomas Davin Fahey, Timothy P White,
the scooter and the electric motor intervenes to assist her et al. Exercise physiology: Human bioenergetics and its applications.
effort automatically. Number Ed. 2. Mayfield publishing company, 1996.
[4] M. Corno, D Berretta, P. Spagnol, and S.M. Savaresi. Design, control,
This objective is achieved by a control system organized and validation of a charge-sustaining parallel hybrid bicycle. IEEE
around three main elements: a kick detection algorithm, Transactions on Control Systems Technology, PP(99):1–1, 2015.
a velocity reference generation and a closed-loop velocity [5] M Corno, F. Busnelli, P. Lisanti, Alli G., and Savaresi S.M. Sistema
control. The heart of the control system is the kick detection per il controllo del moto di un veicolo a propulsione umana di tipo
algorithm. It determines when to activate the assistance. In impulsivo. Italian national Patent ITMI2041985, 2014.
designing the activation logic extra care has to be devoted to [6] M. Corno, M Giani P., Tanelli, and Savaresi. Human-in-the-loop
guarantee that the normal behavior of the kick sequence is bicycle control via active heart rate regulation. IEEE Transactions
on Control Systems Technology, PP(99):1–1, 2014.
preserved. The paper addresses this issue dividing the kick
event in several subsequent phases. [7] M. Corno, Pierfrancesco Spagnol, and Sergio M. Savaresi. Road slope
estimation in bicycles without torque measurements. In Proceedings
The proposed kick scooter is extensively tested and validated of the IFAC Word Congress, page accepted, 2013.
in a human-in-the-loop setting. The analysis thus shows [8] S.A. Fayazy, W. Nianfeng, S. Lucich, A. Vahidi, and G. Mocko.
the full potential of the proposed vehicle in a real urban Optimal Pacing in a Cycling Time-Trial Considering Cyclist’s Fatigue
Dynamics. In American Control Conference, 2013, pages 6457–6462.
environment. In particular, three aspects are considered: the IEEE, 2013.
number of kicks required to cover a given distance, the
[9] Guanetti J., Formentin S., Corno M., and Savaresi S.M. Optimal
average battery power consumption and the rider’s fatigue as energy management in series hybrid electric bicycles. In Accepted
evaluated by heart rate measurements. The scooter is shown for publication to the Control and Decision Conference, 2015.
to be able to reduce the number of required kicks to half as [10] Dong-Min M., Yves P., Miroslav M., Paolo G., and Jian-Xin S.
well as the rider’s fatigue. Thermal modeling of a bldc motor for a kick scooter. In Industrial
Electronics (ISIE), 2012 IEEE International Symposium on, pages
Additional work is being devoted to the complete integration 764–769. IEEE, 2012.
of the all-in-the-wheel kit so to avoid the need of the [11] P. Spagnol, G. Alli, C. Spelta, P. Lisanti, F. Todeschini, S.M. Savaresi,
bluetooth brake switch connection. and A. Morelli. A full hybrid electric bike: How to increase human
efficiency. In American Control Conference (ACC), 2012, pages 2761–
2766, 2012.
45
[12] P. Spagnol, M. Corno, R. Mura, and S. M. Savaresi. Self-sustaining
No assistance strategy for a hybrid electric bike. In American Control Conference
40 Assistance 1%
Assistance 3% (ACC), 2013, pages 3479–3484, 2013.
35 Assistance 6%
Assistance 40% [13] P. Spagnol, M. Corno, and S. M. Savaresi. Pedaling torque recon-
30
Assistance 70% struction for half pedaling sensor. In European Control Conference
J HR [bpm 2]

25 Assistance 100% (ECC), pages 275–280, July 2013.


20 [14] P Spagnol, S Rossi, and Sergio M Savaresi. Kalman filter soc
estimation for li-ion batteries. In Control Applications (CCA), 2011
15 IEEE International Conference on, pages 587–592. IEEE, 2011.
10
[15] Nianfeng Wan, S.A Fayazi, H. Saeidi, and A Vahidi. Optimal power
5 management of an electric bicycle based on terrain preview and
considering human fatigue dynamics. In American Control Conference
0 (ACC), 2014, pages 3462–3467, June 2014.
0 1 2 3 4 5 6 7
Consumption [Wh/km] [16] P.A. Watterson. An electric assist bicycle drive with automatic con-
tinuously variable transmission. In Electrical Machines and Systems,
Fig. 12. Rider fatigue index for different assistance levels. 2008. ICEMS 2008. International Conference on, pages 2992–2997.
IEEE, 2008.

5295

You might also like