Design Principles for Energy Efficient

Legged Locomotion &

Implementation on the MIT Cheetah
Robot
Sangok Seok Et al. (2013)

Presented by:
Rejith Nair Rajeev,
CUSAT
 CONTENTS

1 Introduction

Design principles &

2 Implementation

3 Experiment

4 Conclusions and future work

2
I. Introduction
This paper presents :

a. The design principles for developing highly efficient legged robots,

b. The implementation of the principles in the MIT Cheetah, and
c. The analysis of the high-speed trotting experimental results

• The most widely used criterion for energy efficiency of legged

locomotion is total cost of transport:

TCoT =P/WV
• A low TCoT value is always preferred
3
I. Introduction

4
• TCoT cannot be considered as a perfect approach

• As it is independent from energy dissipations such as joule heating

of coils in electromagnetic actuators and transmission loss in the
gear system, frictional losses etc.

• So, such approach provides a incomplete solution for optimum

efficiency

For developing more energy efficient robots, this paper introduces

design principles for achieving high energy efficiency by analyzing
energy loss throughout the entire system:

◺ Actuator loss,
◺ Transmission loss, and
◺ Interaction loss
5
a b c
6
II. Design principles & Implementation
PRINCIPLE 1 for Ej : High Torque Density Motor

• This principle directly concerns with the Joule heating of the EM

motor by reducing the required electric current to provide torque for
the locomotion.

• If we ignore the variations of the heat dissipation characteristics, the

continuous torque is directly related to motor constant, so as torque
increases, value of motor constant increases (and energy loss in
actuator decreases)

Where,
7
IMPLEMENTATION 1: Large Gap Radius Motor

While maintaining the mass of

the motor constant, the torque
density of the motor can be
improved by increasing the gap
radius: the radius of the gap
between the motor stator
windings and the permanent
magnets on the rotor

8
PRINCIPLE 2 for Ej : Energy Regeneration

• The principal used in electric cars for regenerative braking can be

utilized here to recover the energy lost during the negative work
during the Heel Strike phase of the legs

• The Heel strike phase is considered a negative work as the ground

opposes the swing of legs to move the robot forward.
9
IMPLEMENTATION 2: Motor Drive Electronics Design for Regeneration

• When energizing the motor, the motor driver bucks down the battery
supply voltage to the desired phase voltages of the brushless motor.
• In regenerative mode, the driver acts as a boost converter to step up
the motor phase voltages to more than 100 VDC and recharge the
batteries.
• To verify the effective regeneration during negative work, a simple
experiment was performed:
o To simulate the impact that occurs during running, the robot was lifted
and dropped onto one leg.
o The data from the experiment, shown confirms that the batteries were
indeed recharged. It was found that 63% of the negative mechanical work
done by the motors was recovered by the batteries in the experiment.

10
(a) The voltage of the battery line shows a voltage spike from the boosted
motor voltage
(b) Current flowing out of the battery decreases.
(c) This figure shows the mechanical work done by the motor + the amount of
Joule heating in the motor windings = the power consumption at the
batteries.
11
PRINCIPLE 3 for Ef : Low Impedance Mechanical Transmission

• Legged locomotion requires highly dynamic interactions with the

ground, and the transmission impedance will play a significant role
in the overall dynamics of the robot.

• Employment of gear transmission increases gear friction, actuator

mass, and leg impedance, but it reduces the Ej by increasing the
overall torque density.

• In this experiment they will use a smaller gear train and will not be
considering the joule heating from the transmission design, to
minimize the transmission impedance in order to maximize both the
transmission efficiency and the impedance control capability.

12
IMPLEMENTATION 3: Low Gear Ratio Transmission

• To minimize the losses associated with cascading gear loss, the number
of gear stages is restricted to only one stage in each motor.

• Increasing the number of gear train stages significantly increases

friction, overall inertia of the gears, and complexity of the structure.

• The MIT Cheetah robot uses a

custom designed single stage of
planetary gearing with a gear
ratio of 5.8:1 on each motor

13
PRINCIPLE 4 for Ei : Low Inertia Leg

• Employing low leg inertia is critical for high-speed running.

• Low leg inertia can reduce the torque requirements during the
swing phase. Another benefit of light legs is to reduce impact loss at
touch down at every step.

*∆E, is the impact energy

*v + and v − are the
velocities of before
and after impact
*m is the impact mass

Equation suggests two ways to mitigate impact losses by:

• minimizing the change in velocity at impact (control the retraction
speed of the leg before impact on the ground) 14
• reducing the impact mass
IMPLEMENTATION 4.1: Bio-tensegrity Leg

• They hypothesized that the bones of the animals legs carry mostly
compressive loads while the muscles, tendons and ligaments carry
the tensile loads.

• To achieve this same tendon-bone co-

location architecture, a tendon inspired
material (polyurethane foam and resin) was
integrated into the design of the robots leg.

• Experiments show that this architecture reduces the stress

experienced during stride by up to 59%
15
IMPLEMENTATION 4.2: Dual Coaxial Motor Design

Observations of biological running animals such as horses show that

most of the musculature in the legs is concentrated proximal to the
shoulder/hip of the animal.

Based on the inspiration from this

observation, to minimize inertia the MIT
cheetah robot was designed with two
motors placed co-axially in the shoulder:
one motor actuates the shoulder/hip
joint while the other actuates the knee
through the use of a belt.

16
IMPLEMENTATION 4.3: Differential Actuated Spine

• A switchable spine mechanism was implemented in the design of

the MIT Cheetah to explore the effect of the spine in the dynamics
of running.
• The spine is made up of rings of polyurethane rubber sandwiched
by spine vertebrae segments.
• In the actuated or passive mode, the polyurethane rubber disks can
store elastic energy and return it during galloping.

17
• With the differential actuated spine, the robot can actively arch its
spine up and down and cover more ground in one stride that it
would otherwise with a rigid back.

• When the front legs hit the ground, the spine is also able to absorb
some of the impact and store it as a bending force in the urethane
rubber rings that bring the front legs forward.

18
The motions of the two rear legs are coupled to the motion of the spine
through this differential. Through the innovative use of a differential,
the spine is actuated without the necessity of an extra actuator.

19
20
III. Experiment
• Until now all the principles that have been implemented in the
robot to minimize the losses have been discussed and through
these implementations we got the TCoT value of 0.5

• The aim of this experiment is to identify the most dominant loss

among the 3 losses

• Then further principles can be implemented to minimize this loss

so the efficiency of the robot can be increased further

• So the TCoT value can be reduced even further from 0.5

21
A. Control Algorithm

• This algorithm consists of: a gait-

trajectory generator and pattern
modulator

• The trajectory consists of two

parts: swing phase and stance
phase.

◺ swing time : 250 ms,

◺ stance time : 850 ms 60 ms

22
• The pattern modulator can
describe various animal gait
patterns by synchronizing
individual legs to a specific
pattern with a target speed;
for the test in this paper, a trot
gait is gradually transitioned
to galloping/running.
• The chosen gait pattern was triggered when the left front foot of the
robot touched the ground and generated one stride.

• This process was repeated while the robot was running, and it
successfully modulated the gait pattern over different speeds
23
B. Experimental Setup

• An experimental setup was constructed on a treadmill, the

maximum speed of the treadmill was 6 m/s
• A linear guiderail system is mounted above the treadmill. A pair of
rail sliders is attached to the robot via metal bars and a rod that goes
through its center of mass

24
25
C. Experimental Results
• They measured the voltage and the current data of each motor and
the treadmill speed.
• High speed video was captured at 500 fps by a high-speed camera

26
• Mechanical power, Joule heating and total power were calculated
using these equations:

• Power consumption measured at battery matches up with the sum

of Mechanical Power + Joule Heating, so the power consumption at
the battery is not plotted
• The power consumption by the motor drivers is included in Joule
heating

27
The change in power consumption at 6 m/s is shown:

28
• It can be seen that the mechanical power(black line) showed negative
values but it was regenerated during locomotion. This regenerated
power was supposed to charge the battery.

• But it is seen that the total power briefly reaches zero & did not go
negative, which means that the generated energy when a pair of legs
did negative work was fed to the other pair of legs to do positive
work.

• Most of the times regenerated energy goes to other joint actuators

and only little energy recharges the batteries.

• But, in several exceptional cases, such as during landing from a jump,

we expect that a significant amount of energy will charge the
batteries.
29
This table shows several attributes of the average power consumption
at 6 m/s. Of the 973 W consumed, 739 W is lost in Joule heating, while
234 W is consumed by the net mechanical work, which is dissipated
through gear friction, joint friction, and interaction loss.

30
IV. Conclusion
• Based on the energy flow of a locomotive system, four major design
principles were highlighted: high torque-density motors, an energy
regenerative electronic system, low loss transmission, and low leg
inertia.

• The MIT Cheetah achieved a cost of transport of 0.5 that rivals

running animals and is significantly lower than other running
robots.

• From the experimental results, specifically analysis of energy loss in

the components of the system, we learned that Joule heating is the
major culprit for power loss during locomotion.
31
• In order to reduce the Joule heating from electric motor, we have
developed a new custom-designed 3-phase motor focused on
reducing heat loss and maximizing torque density.

• The preliminary experiment data

shows that it has 3 times the
saturation (peak) torque of the
current motor.

• With this, it is expected that the

TCoT will be reduced to 0.25,
which is superior to efficiency of
biological runners and similar to
that of fliers in nature.
32
Thank You

33