Design, Fabrication and Control of a Two-Legged Walking Robot

Yasar Ayaz, Bilal Afzal, Mannan Saeed, Saeed-ur-Rehman

Department of Mechatronics Engineering,

College of EBrME, NUST, Rawalpindi, Pakistan.
Email: [email protected],[email protected]

Abstract and implementation of feedback control. The paper also

employs a simple technique for handling the instability
A 4-DOF (Degrees of Freedom) two-legged robot named problem often encountered in such robots; as a result the
NUSTBOT-I has been designed and built. The paper robot does not fall when it bends to one side with a shift
contains a description of the design and working of this in its centre of gravity.
robot, supported by mathematical analysis of its
structure. The results obtained from this analysis have This paper is organized as follows: Section 2 describes
been applied for tracing the trajectory of the centre of the main design ideas which lay the foundation for proper
gravity of the robot, as it walks. One of the main design functioning of the robot, system modeling is presented in
objectives was the simplicity of its control algorithm. The Section 3, Section 4 highlights the salient features of the
robot shifis its centre of gravity by bending the upper part control strategy while COG trajectory of the robot is
of its body sideways. It can perform basic walking and traced in Section 5 and Section 6 concludes the paper by
turning operations and is programmable for different summarizing the results.
applications even by users who do not have any technical
knowledge about bipedal walking.
2. Conceptual Model
1. Introduction While designing NUSTBOT-1, the main objectives were to
implement the basic motions of walking and turning but at
Two legged walk is a trait that distinguishes “humanoids the same time keeping the structure simple, lightweight
or bipeds” from other kinds of robots. In our paper we and easy to handle. Thus, for shifting of the body weight
have described the construction and working of (i.e. COG shifting) on either foot, an inverted pendulum
NUSTBOT-1, a 4-DOF two-legged robot developed at has been used. Batteries, which constitute 1/4 of the body
college of E&ME. Unlike other such robots r1-31 weight, are used as bob of the pendulum. When the
NUSTBOT-1 can easily be programmed by users who pendulum (or shifting arm as we call it) falls to one side,
have no technical knowledge about bipedal walking. It is the center of gravity of.the body suddenly shifts over to the
low cost, light-weight and user-friendly. The basic foot on that side resulting in net torque acting on the body.
walking mechanism begins with the robot shifting its Under the action of this torque, the body bends in the
centre of gravity (COG) over one foot by bending its direction of the torque (Fig.1) until a mechanical support
body to one side, thus lifting the other foot -off the attached to the foot comes into contact with the leg. This
ground. Standing on this one foot, the robot now moves action introduces damped oscillations, of small amplitude,
the free leg forward and then shifts its centre of gravity in the upper body, which die out quickly and then the body
over to this forward extended foot, followed by the joint becomes balanced on that foot.
movements to straighten its body, thus completing one
step forward. The paper contains a thorough discussion As obvious from Fig.1, the forward or backward motion
on the design and working of our biped robot, supported of the foot is possible by rotating the leg at the hip (e2 or
by mathematical analysis of its structure and e3).In order to ensure that when weight is shifted from
determination of its centre of gravity. The results one foot to the other during the stride, the foot falls flat
obtained from this analysis have been applied for tracing on the ground (no matter what the step size), the standard
the trajectory of the centre of gravity of the robot, as it four bar parallel mechanism is used for each leg (Fig. I).
walks, thus clearly defining the extremes of movements Thus when one foot lies flat on the ground, the other foot
over which the robot would be stable. We have also automatically remains parallel to the ground even when
highlighted the basic ideas used in the mechanical design stretched forward.

Fourth International Workshop on Robot Motion and Control, June 17-20,2004 73

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.
 3. System Modeling
The basic technique used in the design of the mechanical
structure is to keep it symmetric in the weight shifting
plane about the mean position of the shifting arm.This
assures that at mean position of the arm the weight of the
robot is equally distributed over both feet. Also that the
arm is capable of shifting weight to either foot to the
same limit. The major materials used are Aluminium for
the body structure, Oil Hard Steel for gear systems and
Mild Steel for couplings.

The location of center of gravity (COG) of a body in a

uniform gravitational field can be determined by dividing
the body into parts of geometrically simple shapes of
uniform mass density such that their individual COGS can
he located geometrically. Then the mass (mi) and the
distance of the COG along an arbitrary axis (r,)for each
of these parts (where i is the index of the part) is
determined. The position of the COG of the whole body
can be determined [4] by using the formula:

m rcoc= 1 miri
i=l

where m is the mass of the body and rcoc is the distance

of the COG of the body along the arbitrary axis. This
procedure has to he carried out separately for each of the
three axes for three dimensional analysis.
(h)
Fig .l. (a) Robot Structure (b) Location of DOFs
To apply this procedure on NUSTBOT-1 we divided it
To make NUSTBOT-1 capable of taking turns we into 6 distinct structural units. These are feet (g,h),
decided to split the plate joining the two legs into two, legs(e,t), turning gear system(c), weight shifting gear
with a pin joining them. On this pin one leg can rotate system(b), body plates(d) and weight shifting arm(a).
First the masses and COG locations for all these parts are
(84) independent of the rest of the body up to a certain determined separately using (1). As an example, the
angle constrained by the mechanical design. Thus by calculation for the location of the COG of the shifting
shifting the weight onto one foot first, then rotating the arm (Fig.2.) in the vertical axis is given below:
other leg, then shifting the weight onto the other foot and
rotating the leg off the ground in the same direction, the
robot can take small turns.

Since it is important for us to know only whether a

particular joint has reached a certain position yet or not, it
was decided to implement the control system using limit
switches with low speed DC motors. The DC motors
were selected keeping in view their relatively higher
torque as compared to stepper motors of a similar size.
Also, the DC motors of our requirement were rather light
weight which assisted in keeping the robot lighter, since
each motor has to lift the rest of the motors in one or
more of its movements during the walking sequence. Fig.2. Weight shifting arm

74

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.
 ma= 0.796 kg For y-axis
5
mar,= miri
i=l ,=o
r,,= 0.197111
= (0.4929 + 0.157 COS e,) I 2.925 (3)
Using this procedure we can find the individual masses
and COG locations for each of the structural units and where 0, for (2) and (3) is the rotation angle of the
then determine their coordinates in the global axis system shifting arm measured from its mean position, and
for the robot. The origin for this axis system lies at the positive in the anticlockwise direction when the robot is
hack outer comer of the right foot of the robot. The viewed from the front. x and y give COG coordinates in x
allocation of direction for each axis can be seen from and y axes in the global axis system.
Fig.3 and 4 below as we determine the COG position for
the robot. By applying a similar approach COG location along z-
axis is determined as:
Using Fig.3, mass of the whole robot is determined as:
z= (0.086 + 0.495 Sin&+0.043 Sin&) 12.925 (4)
m= mi where the z-coordinate will always he determined
i=a forwards from the posterior end of the back foot using
m = 2.925 kg (4). hand b a r e shown in Fig.4.

Fig.4. Calculating robot COG in z-axis

Putting 0, = @ b = @f= 0 in (2),(3) & (4),

(x,y,z)= (0.0753m, 0.2222m,O.O29m)

This is the location of the COG of the robot at the mean

position as shown in Fig.3. For weight shifting we shift
the arm 45" in the desired direction.

For 0, = -45", x = 0.1 13m => on left foot,

y= 0.206m,
Fig.3. Calculating robot COG in xy-axes Net Torque = 0.23 Nm (CW)

Using (1) on Fig.3, for x-axis, For 0, = 4 9 , x = 0.037m => on right foot,
y = 0.206m,
h Net Torque = 0.23 Nm (CCW)
mx= C mixi
i=a Under this torque the body of the robot falls to a side by
rotating over the free ankle joint. Since during this falling
x = (0.22-0.157 Sin 8,)12.925 (2) motion, the body of the robot behaves like an inverted

75

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.
 pendulum swinging on the ankle joint, equations for 6, is restricted to rotating only 10" in one turning step.
COG location during this motion can be trivially M(0) and M ( I ) are object matrices of the right foot before
determined. and after rotation respectively as shown in Fig. 5 .

x = 0.075 - (Ss / ISsl) 0.03 - 0.206 Sin S,

--
(5)
y = 0.206 COSSs (6) 4. Control Strategy
6 is determined from the ankle joint as shown in Fig.7
later on and is taken to he positive in anticlockwise
direction.
<y-
The robot is allowed to fall freely for Ss = 6.21' either
way after which the mechanical support attached to the I J

foot comes into contact with the body and stabilizes the
falling system with the COG balanced at the center of one
foot and the other foot lifted off the ground.

For the turning feature, the new coordinates for the foot
when it turns can be determined using spatial
transformations 151. The axis of rotation is along the y-
axis of an arbitrary frame whose origin is located at the
geometrical centre of the two feet of the robot (Point 0 in
Fig.5.).

IC%--

Fig.5. Coordinate transformation during a 10' left

turning

e
Ice,= hmnc

End d Sequenoe

Fig.6. Working Sequence for taking one step forward

(WALK-LEFTSEQUENCE)

Po 212
0
Z
0
j
A total of four DC motors are required to control
NUSTBOT-I i.e, two for leg movements, one for turning
and another for weight shifting. Each of these motors
runs on 12V DC in our design and draws a maximum

76

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.
 current of 0.2A. To suppolt this requirement two 6V DC,
1.3Ah batteries have been used in series as the power
source. .. ..
i.....
For hi-directional control of each motor the control i I
I
system has been implemented using H-bridge circuits.
For this purpose P I C 0 2 9 8 has been used which is a dual i
H-bridge IC from Texas Instruments. The control
inputs for the H-bridge come directly from an
i :
ATMEL89c51 microcontroller which is the logic device i
in our circuit. The feedback regarding the state of the iI .
joints is provided to the microcontroller by limit
switches. There are three limit switches to monitor the
position of each actuated joint, two for extremes and one
for the mean (or home) position.

The software programming of the controller has been

done using a layered approach [6].These layers introduce
different level of control to the users at each layer. The
bottom layer deals with the motors directly. It canies out
control of the 'status' (odoff) and 'direction' of the
motors in two separate registers, thus making it possible
to implement motor control techniques like PWM
without programming complexity. The medium layer
utilizes the bottom level functions to implement complete
sequences of movements in subroutines. For example, a
routine named WALK-LEITSEQUENCE has been
implemented that makes the robot walk a step forward by
standing on the right foot and stretching forward with the
left one. The top layer is for users who do not have
knowledge about the technical functioning of the two
legged walking machine and want to program it for some Fig.7. Robot COG trajectory in xy-plane
application. These users can use the routines exported by
the medium layer and just give instructions like walk For terminal positions:
right, turn left etc to the robot.
e,= -6.27' => a= 0.1275m =>on left foot
The algorithm has been implemented using assembly & y=0.204m
language. The control system is a sequential 'bang-bang' e,= 6.27' =>E 0.0225m =>on right foot
type control system which is event driven. The control & pO.204m
software increases the flexibility, making the user capable
of using a variety of techniques to control NUSTBOT-I. We see that at either terminal point the robot balances
Fig.6 shows a typical walk forward sequence in which the itself with its COG placed at the centre of one foot.
robot uses its right leg as the weight bearing one while
stretching forward with the left.

5. COG Trajectory
Trajectory of the COG of the robot in the area marked
STABLE in Fig.7 is governed by (2) & (3) while the rest
of the motion is governed by ( 5 ) & (6).

Table.1. COG trajectoryfor walking one step forward

77

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.
 Table.] gives the trajectory of the robot’s COG while observed that for more stable walking 4 should be kept
walking a step forward by standing on the left foot as significantly larger than &.
shown in F i g 3 below: Here, d=0.184 ( S i n 4 + Sin@) is
the distance of back end of front foot from origin in z-
axis and z, is the distance of COG along z-axis from the 6. Summary
original origin location for c a e s (v) and (vi) when the
foot marking the origin has moved forward. By virtue of a simple design and a layered control
software, we have accomplished our goal of making an
easy to handle and program, user-friendly two-legged
walking robot. The use of a mechanical suppolt for
balancing of the robot and limit switches for joint
position feedback have provided easy solutions to
complex problems. However, the turning mechanism
restricts NUSTBOT-1 to taking turns of multiples of 10”
only. Future work can be carried out to make the turning
motion of the robot more flexible.

Acknowledgements
Special thanks to Dr. Khalid Munawar of the Department
of Electrical Engineering for for his keen interest in our
work and for letting us use his labs for several of the
experiments during our research.

References

[ I ] Jin-Woo Lee,Ki-Hoon Lee, Ill-Woo Park and Jun-Ho

Oh, ‘‘Walking and Control of Autonomous Biped
Robot”, IFAC .,,,Workshop on Mobile Robot
Technology, pp:179-183, Jejudo Island, KOREA,
May 20-23, 2001.

__1 121 Jin-Woo Lee , Sung - H w n Kim, and Jun-Ho Oh,

“Design of four joints and three links biped robot and
its gaits”,lnt. con$ on Intelligence Autonornus
(ii) Systems, pp. 295-300, Italy, Venice, 2000.7.

[3] Sung-Hoon Kim, Jnn-Ho Oh, and Ki-Hoon Lee,

“Design of 4 Joints 3 Links Biped Robot and Its
Gaits”, IEEWRSJ International Conference on
Intelligent Robots and Systems, pp. 1155-1160 ,
(0 October 30 - November 5,2000
Fig.8. Robot COG trajectory along z-axi
141 www.citvcolleriate.com/centre of mass.htm
One can see from Fig.8. that during the entire walk, the
COG remains restricted within the polygon defined by 151 John J. Craig, Infroduction to Robotics (Mechanics
the robot’s feet. For the sequence shown in Fig.8, the and Control), Addison Wesley Longman, Inc, 1999.
robot’s step size is 0.078m with 4 = 25’ and @ = 0”.
Different step sizes can be chosen for the robot by [6] R. Brooks, “A Robust layered Control System for a
choosing different values for and @. However, it is Mobile Robot”, IEEE Journal of Robotics and
Automation, 2 (l), 1986.

78

Authorized licensed use limited to: NUST School of Electrical Engineering and Computer Science (SEECS). Downloaded on March 20,2024 at 09:26:50 UTC from IEEE Xplore. Restrictions apply.