Upload
vuongtuong
View
215
Download
1
Embed Size (px)
Citation preview
Int j simul model 10 (2011) 4, 177-190
ISSN 1726-4529 Professional paper
DOI:10.2507/IJSIMM10(4)2.186 177
OPTIMUM STABLE GAIT PLANNING FOR AN 8 LINK
BIPED ROBOT USING SIMULATED ANNEALING
Sudheer, A. P.*; Vijayakumar, R.
* & Mohandas, K. P.
**
* Department of Mechanical Engineering, NIT Calicut, India
** Department of Electrical Engineering, NIT Calicut, India
E-Mail: [email protected]
Abstract
Biped or humanoid robots should have a higher level of autonomy to achieve different tasks
in service and manufacturing sectors. Biped locomotion is a complex dynamical task because
of the intermittent interaction with its environment and the ground through its feet. Most of
the previous gait planning efforts were made by complex and time consuming algorithms with
maximum one limb in the upper body for controlling the gaits. This work presents the
kinematic and dynamic modelling and simulation of 8 link biped robot walking with two
limbs in the upper body. Optimum cycloidal gait trajectory is simulated and compared
between the static and dynamic walking cases. Easy and comparatively less time consuming
simulated annealing algorithm is used for the optimization. Zero Moment Point criteria is
used for the stability analysis. (Received in December 2010, accepted in August 2011. This paper was with the authors 2 months for 1 revision.)
Key Words: Gait Planning, Sagittal Plane, Zero Moment Point, Simulated Annealing,
Degrees of Freedom
1. INTRODUCTION
One of the main objectives of humanoid robotics is to create autonomous systems for
imitating humans at a wide variety of dynamic, complex and novel environments. Experts
from various fields of science, engineering and linguistics have tried to combine their efforts
to create a robot as human-like as possible. Although this field has witnessed tremendous
progress in recent years, realistic, periodic and rhythmic gait generation is not yet achieved
due to the limitations in kinematics, dynamics and control techniques.
Biped robot walking consists of two stages, namely single support phase in which one
foot is in contact and double support phase in which both feet are in contact with the ground.
In a complete walking cycle, stability has to be analysed in both the above phases during
static and dynamic conditions. Position of Center of Gravity (COG) and Zero Moment Point
(ZMP) are two important criteria for analyzing the static and dynamic stability of legged
robots. Qiang Huang et al. [1] interpreted ZMP, as that point on the ground about which the
sum of all moments of the active forces is equal to zero. This concept adopts Vukobratovic et
al.’s [2] claim that the ZMP has been vaguely related to the ground surface, regardless of
referring to the support foot polygon of the legged robot. Researchers have been working on
gait planning and executing gaits for humanoid or biped robots based on genetic algorithms
(GA), artificial neural networks (ANN) and fuzzy logic control (FLC) and other conventional
algorithms [3-12] considering stability mostly only in the sagittal plane. Goswami Dip et al.
[3] optimized the walking gaits using genetic algorithm (GA) by considering the Zero
Moment Point (ZMP) and walking speed. The optimal walking gaits are also experimentally
realized on the biped robot also. Maitray Shrivastava et al. [4] used a trajectory generation
method using GA for a 8 degrees of freedom (DOF) robot that can walk on flat terrain and
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
178
climb stairs with deformation at the sole. This method incorporates the dynamics of an actual
8 DOF robot to find the most energy optimal gait.
Jih-Gau Juang [5] developed a technique based on artificial intelligence for a 5 DOF biped
robot for the trajectory control. A three layered ANN is used as a controller. It provides the
control signal in each stage of walking gait. Chenglong Fu and Ken Chen [6] proposed
optimum walking control for humanoid stair climbing, which consists of a stair climbing gait
and a sensory control strategy. The climbing gait is considered as feed forward control and is
modeled to satisfy the environmental, kinematic and ZMP constraints. J. B. Goncalves and D.
E. Zampieri [7] used a recurrent neural network to determine the trunk motion for a biped
walking by considering the ZMP criterion. Sathoshi Ito et al. [8] introduced a ZMP feedback
control algorithm for 4 DOF biped balance in the lateral stepping motion. Stability of the
stepping motion is examined analytically and through computer simulation. Effectiveness of
this method is demonstrated by biped robot experiment during the lateral stepping motion. N.
Bigdeh et al. [9] developed a method for dynamic modelling of a five link robot based on
neural network. This method considerably reduces the complexities in solving the dynamic
model equations. Jong Hyeon Park and Moosung Choi [10] proposed a method that
minimizes the energy consumption in the locomotion of a 7 DOF biped robot. A real coded
GA is employed in order to search for the locomotion pattern and optimum center of mass
locations of links of the biped robot. Rahul Kumar Jha et al. [11] modelled the gait generation
problem for the step climbing using fuzzy logic controller and its rule base is optimized
offline using GA in the simulation of the gait trajectory. Panda Ranga Vundavilli and Dilip
Kumar Pratihar [12] proposed a method for the design of a neural network based gait planner
for a two legged robot negotiating uneven terrains. Genetic algorithm is used to provide
training off-line to the gait planner and the trajectory is simulated. These authors have
extended their work for biped motion up and down the sloping surfaces and ditch crossing by
combining soft computing techniques like GA, FLC and ANN.
In this paper, Simulated Annealing (SA), an intelligent random-search technique is used
for getting the optimum stable gait trajectory in both static and dynamic conditions of biped
robot by carefully controlling the rate of control parameter. An 8 link biped robot with
frictionless joints is used in this work for the stability analysis and simulation. Static and
dynamic stability cases are considered separately, for the optimum gait planning and analysis.
Peak error in ZMP is minimized for getting the static and dynamic stability when biped gait is
generating one walking step. Simulated annealing algorithm gives the solution within a short
time for this problem. To the knowledge of the authors, no work based on simulated
annealing has been reported in the area of humanoid or biped robots. This paper is structured
as follows. In Section 2, both kinematic and dynamic modelling are described. The
fundamental theory of the center of mass (COM) and ZMP, in single and double support
phases are given in Section 3. Formulation of the problem and simulated annealing algorithm
are shown in Section 4 and 5 respectively. Section 6 deals with the optimal gait trajectory for
stepping motion along with the computer simulation results. Finally Section 7 presents the
concluding remarks with outlook.
2. MODELLING OF BIPED ROBOT
An important part of robot simulation is the treatment of robot kinematics and dynamics. The
advances in humanoid robotics and the increased usage of biped or humanoid robots have
raised the need for computer simulation of robots among the aims of which are the design of
new robots, task planning of existing robots, performance evaluation and cycle time
estimation. Modelling of robots includes both kinematic and dynamic modelling.
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
179
2.1 Kinematic modelling
There are different methods available for the kinematic modelling and analysis. In general, the
inverse kinematic problem can be solved by various methods such as screw algebra (Kohli,
1975), dual matrices (Denavit, 1956), iterative (Uicker, 1964), geometric approach (Lee,
1984) and decoupling of position and orientation (Pieper, 1968). Geometric approach is used
for deriving the inverse kinematic equations in this work. Biped robot model with mass of
each link, assumed to be concentrated at different position for the gait synthesis, is as shown
in Fig. 1. The coordinates of hip joint J3 and ankle joint J6 of the swing leg are (xh, yh) and (xe,
ye) respectively. Forward and inverse kinematic equations are formed for the motion and
stability analysis, based on the geometry of the biped. Inverse kinematic equations derived
based on the hip position for the single support phase (SSP) and double support phase (DSP)
are listed below. Joint positions θ2 & θ3 are same for both phases and θ1 & θ8 are calculated
from the simple geometry.
222
23
22
22
2
2arccosarctan
2hh
hh
h
h
yxl
llyx
x
y (1)
223
22
23
22
3
2arccosarctan
hh
hh
h
h
yxl
llyx
y
x (2)
226
27
26
22
6
)(2
)(arccosarctan
hhe
hhe
h
he
yxxl
llyxx
y
xx (3)
227
26
27
22
7
)(2
)(arccosarctan
2hhe
hhe
he
h
yxxl
llyxx
xx
y (4)
Figure 1: Biped robot model.
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
180
Inverse Kinematic Eqs. (1) to (6) gives the instantaneous angular positions of various
links during the stepping motion. The motion of a biped robot comprises of time-functions of
angular positions and velocities of the joints and links of the robot. The straight forward
approach is to generate the joint time trajectories by solving inverse kinematics, to maintain
physical stability of the biped. Eqs. (5) and (6) are used for finding joint parameters of the
sixth and seventh links in SSP.
226
27
26
22
6
)()(2
)()(arccosarctan
lyyxxl
lllyyxx
lyy
xx
ehhe
ehhe
eh
he (5)
227
26
27
22
7
)()(2
)()(arccosarctan
2 lyyxxl
lllyyxx
xx
lyy
ehhe
ehhe
he
eh (6)
Joint angles θ4 and θ5 are obtained directly from the simulation by using the simulated
annealing algorithm in both static and dynamic cases. Table I shows the parameters of various
links of the biped robot. The moment of inertia values listed in the table are used for the
dynamic stability analysis.
Table I: Parameters of biped robot links.
Link (i) Mass (mi)
[kg]
Length (li)
[m]
Center of mass
from the joint (ri)
[m]
Moment of
inertia (Ii)
×10-2
[kgm2]
1 0.06 0.05 & a = 0.05 0.03 0.0188
2 1.5 0.30 0.24 2.3400
3 2.5 0.35 0.28 5.3084
4 8 0.25 0.2 8.6666
5 2 0.15 0.12 0.7800
6 2.5 0.35 0.28 5.3084
7 1.5 0.30 0.24 2.3400
8 0.06 0.05 & a = 0.05 0.03 0.0188
2.2 Dynamic modelling
Dynamic model of a robot is useful for the computation of torque and forces required for
execution of a typical work cycle, which is vital information for the design of links, joints
drives and actuators. A general dynamic model for biped walking related to the joint
coordinates vector and joint torque vector without considering the friction and other external
disturbances is given in Eq. (7).
DGCM )(),()( (7)
where, M is the 8×8 inertia matrix, C = )()( hB is the 8×8 Coriolis and centrifugal matrix
and G = Cg(θ) is the 8×1gravity vector, D is the 8×8 coefficient matrix of joint torques.
M(θ) = [qij cos(θi – θj)]; B(θ) = [qij sin(θi – θj)]; C = – [diag(hi)];
θ = [θ1 θ2 θ3 θ4 θ5 θ6 θ7 θ8]T; Th 2
82
72
62
52
42
32
22
1)( ;
g(θ) = [sin θ1 sin θ2 sin θ3 sin θ4 sin θ5 sin θ6 sin θ7 sin θ8]T;
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
181
τ = [τ1 τ2 τ3 τ4 τ5 τ6 τ7 τ8]T and
1 0 0 0 0 0 0 0
0 1 -1 0 0 0 0 0
0 0 1 -1 0 0 0 0
0 0 0 1 -1 0 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 1 -1 0
0 0 0 0 0 0 1 -1
0 0 0 0 0 0 0 1
τi is the torque at the ith
joint θi and
.
i are the angular position and velocities of link i. The
parameters qij and hi are constants derived by Langrangian equations of motion.
3. STABILITY CRITERIA
In static gait planning problems, biped robot is stable if the projection of COM falls within the
convex hull of the foot support polygon. In dynamic locomotion, link acceleration, inertial
forces, and ground reaction force are also to be considered and the ZMP should be within the
convex hull of the foot polygon for satisfying the stability criterion. The dynamic locomotion
is highly nonlinear and difficult to analyze in real environment. The condition in the static and
dynamic stability of the biped along the sagittal plane during the single support phase is that
the location of the ZMP (Xzmp, 0, 0) must be inside the convex hull of the supporting foot.
Therefore, Xmin < Xzmp < Xmax, based on the biped foot dimension. In double support phase
ZMP or projection of COM should lie within the convex hull of support polygon formed by
left and right foot. In the case of static stability analysis Xzmp equation contains only mass and
the displacement terms. The Xzmp can be expressed as:
Xzmp =
8
1
8
1
8
1
8
1
)(
)(
iii
i iiiiiii
iii
gym
wIyxmxgym
, Yzmp = 0 and Zzmp = 0 (8)
4. PROBLEM FORMULATION
Mathematical interpolation is one of the simplest methods for providing suitable gait
trajectory in accordance to the given boundary conditions. Generally the trajectory of the
swing foot of human is a cycloidal profile in normal walking [5]. Cycloidal profile reduces
the effects of sudden acceleration at the beginning and deceleration at the end during the gait
generation. In this simulation, swing foot trajectory is taken as cycloid and the hip joint
trajectory is taken as a modified cycloid. The modified Cartesian cycloid equation used for the
hip joint motion is given as:
x = rx (θ – sin θ) and y = ry (1 – cos θ) (9)
The objective of this work is to generate an optimal cycloidal gait trajectory by
minimizing peak error in ZMP by changing the hip and upper body link positions of the biped
robot. If the ZMP variations are minimum we can select a minimum foot size so that biped
can move easily without any external and self collisions. One walking step begins with the
heel of the right foot leaving the ground. Single support phase is the predominant portion
D =
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
182
because its duration is longer than double support and transition phases. In this paper, a
complete cycloidal gait trajectory is optimized based on the ZMP value when the hip joint is
moving through number of modified cycloidal paths and also by changing upper body
positions. The objective function is:
Minimize f = Minimize (Max(XzmpError)) = Minimize (Max[|(Xzmp – Xzmp – reference)|]) (10)
where: ,,,,(),,,,,,,( 5544 iiiififihihiyx yyxxfyxrrfXzmpError iw );
with constraints: 0 ≤ rx ≤ 30; 0 ≤ ry ≤ 30; -30 ≤ xhi ≤ 0; 40 ≤ yhi ≤ 70; 0 ≤ θ4i ≤ π/4;
0 ≤ θ4f ≤ π/4; 0 ≤ θ5i ≤ π/4; 0 ≤ θ5f ≤ π/4.
Cartesian and joint positional constraints used in both static and dynamic walking
simulation are in centimeters and in degrees respectively. The variables which define the
modified cycloidal path of hip are rx, ry, xhi, and yhi ; where rx and ry are the span and height
and (xhi ,yhi) is the starting coordinates of cycloid. Other variables that determine the optimal
gait trajectories are θ4i, θ4f, θ5i and θ5f , the initial and final upper body joint variables. These
eight variables determine the optimal gait trajectory of the swing foot with minimum stance
foot size. Origin of the inertial reference is taken as the ZMP reference in this work.
Following assumptions are made for the optimal gait planning.
Swing foot in the first half of the stepping time moves with uniform acceleration and in
the second half moves with uniform deceleration.
During single support phase, stance foot is assumed to remain in flat contact on the
ground.
The coefficient of friction for rotation at joints is zero and the links are perfectly rigid.
The contact between the floor and the foot is smooth and floor for walking is rigid.
5. SIMULATED ANNEALING
Simulated annealing (SA), a random-search technique, was developed by Kirkpatrick et al.
[13]. This is a powerful optimization method inspired by metallurgical annealing process
which exploits an analogy between the way in which a metal cools and freezes into a
minimum energy crystalline structure. SA's major advantage over other methods is an ability
to avoid becoming trapped in local minima. It has been proved that by carefully controlling
the rate of cooling of the temperature, SA can find the global optimum in a reasonable amount
of time. SA algorithm does not need to calculate the gradient descent that is required for most
traditional search algorithms. Various steps involved in the simulated annealing based
optimization are depicted below.
Step 1: Choose a set of initial variables or point Xt, say X1 = (rx, ry, xhi, yhi, θ4i, θ4f, θ5i, θ5f) and
set a starting temperature Ts a sufficiently high value, cooling rate Cr, and set final
temperature Te. The algorithm starts from a randomly chosen point X1 of which the fitness
value of the objective function, E(Xt) is computed.
Step 2: A neighbouring point Xt+1, say X2 gets chosen using Gaussian distribution and E(Xt+1)
is computed.
Step 3: Calculate ∆E = [E(Xt+1) − E(Xt)], if ∆E ≤ 0, i.e., the cost of the neighbouring point is
better than the cost of the point, the new point is accepted. Otherwise the new point is
accepted by a probability p = exp (−∆E / Ts) if the random number r generated in the range
(0, 1) is less than p. Else go to step 2. This completes one iteration of the simulated annealing
procedure.
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
183
Step 4: Calculate the new temperature, Tnew= Ts ×Cr and then check the terminating condition.
The algorithm is terminated when Tnew ≤ Te or a small enough change in fitness value is
obtained. Flow chart of this particular work with the selection of various parameters is given
in Fig. 2.
Figure 2: Flow chart for simulated annealing algorithm.
6. SIMULATION RESULTS AND DISCUSSION
Statically and dynamically stable optimal walking trajectory is of 8 links biped robot is
determined by using simulated annealing algorithm. Present paper gives an idea about the
influence of upper links and hip positions in the biped stable walking. Initialization module
includes the selection of control parameter Ts = 200, cooling rate Cr = 0.95 and the final
control parameter Te = 30 as shown in Fig. 2. These selected values are found to be good for
this problem because the solution is arrived at within a reasonable time. The possibility of
being trapped in a local minimum was overcome with more iterations of simulation within the
0 < r < 1
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
184
required value of fitness. Statically stable walking is analyzed by shifting only the masses of
various limbs of the robot without considering the parameters like acceleration, inertial forces
and other disturbances. But in dynamic stable walking these parameters are also considered in
the simulation. Simulation is carried out on a Core 2 Duo Intel processor using MATLAB.
Nine intermediate points in between starting and ending positions in the cycloidal
trajectory of the swing foot are considered for analyzing the gait. Initially the maximum value
of absolute ZMP error is determined out of all intermediate position ZMP errors in each
cycloidal motion based on modified cycloidal hip trajectory and upper body positions. Then
minimized the peak absolute error in ZMP in the solution space by using the simulated
annealing algorithm, which will give the idea of the required the foot size for an optimum gait
of the biped.
6.1 Static stable biped walking
Statically stable walking pattern is simulated and the kinematic diagram is given in Fig.3. The
required optimum value of peak error in ZMP is fixed in this case, that is when (XzmpError, 0, 0)
≤ (0.3, 0, 0) the simulation is stopped. The period required for one walking step is assumed as
5 s with a step length 20 cm. The simulation is carried out ten times for checking the number
of iterations and time taken for each fitness value. It is found that the time taken in all the
simulations is reasonable. Table II gives an idea of the time taken, peak error and the values
of variables of simulation of biped optimum gait planning when the robot is walking in
statically stable condition. Fig. 3 depicts the optimal gait trajectory of the swing foot of the
biped which satisfies the required static stability criteria. This stick diagram gives the idea of
the influence of the upper body and the hip motions for generating an optimum gait. Angular
positional variations of fifth, sixth and seventh links are more compared to other links as
shown in Fig. 4. That is the angular positional variations of these links help more to bring the
biped in statically stable condition. ZMP error variations and fitness variations are plotted in
Figs. 5 and 6 respectively. From the fitness plot it is clear that simulated annealing algorithm
helps to escape from the local minimum because of the probabilistic hill climbing
characteristics of SA and independency of initial conditions. This characteristics of SA helps
to find global minimum. Table II shows the comparison of parameters in ten set of static
stable simulations. Hip positions in 6th
, 7th
and 8th
simulation are same and reasonably good
with the same value of peak error. In 4th
and 5th
simulation peak error obtained is minimum
with the same variations of upper body and hip positions.
Figure 3: Static walking.
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
185
-40
-30
-20
-10
0
10
20
30
40
50
60
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (S)
Jo
int
An
gle
s(D
eg
ree)
Theta 2
Theta 3
Theta 4
Theta 5
Theta 6
Theta 7
Figure 4: Variations of joint angles. Figure 5: Variation of ZMP.
Table II: Comparison of parameters in statically stable simulation.
Simulation
No.
Minimum
peak error
(< 0.3) [cm]
Time
taken
[s]
No. of
iterations
Variables
hs = [rx, ry, xhi, yhi, θ4i, θ4f, θ5i, θ5f ]
1 0.2851 115.55 439 [ 0.2483 3.0819 -7.7969 62.3894
0.2860 0.2321 0.7569 0.0594 ]
2 0.2851 114.97 436 [ 0.2483 3.0819 -7.7969 62.3894
0.2860 0.2321 0.7569 0.0594 ]
3 0.2479 119.96 459 [ 2.8911 4.9687 -6.7284 60.1678
0.2492 0.3673 0.1114 0.0522 ]
4 0.1931 133.83 524 [ 2.3783 3.1394 -6.3684 60.3379
0.2198 0.2832 0.0305 0.0037 ]
5 0.1931 132.70 536 [ 2.3783 3.1394 -6.3684 60.3379
0.2198 0.2832 0.0305 0.0037 ]
6 0.2969 131.17 511 [ 3.9685 5.0167 -3.3416 62.177
0.0221 0.1370 0.0065 0.4182 ]
7 0.2969 130.42 504 [ 3.9685 5.0167 -3.3416 62.177
0.0221 0.1370 0.0065 0.418 2]
8 0.2969 131.47 515 [ 3.9685 5.0167 -3.3416 62.177
0.0221 0.1370 0.0065 0.4182 ]
9 0.2926 139.08 521 [ 2.5563 3.1811 -4.0167 62.0584
0.0086 0.0309 0.5515 0.4858 ]
10 0.2926 140.09 534 [ 2.5563 3.1811 -4.0167 62.0584
0.0086 0.0309 0.5515 0.4858 ]
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
186
Figure 6: Variation of fitness function.
6.2 Dynamic stable biped walking
Dynamic walking analysis is important when robot is running, jumping or performing other
types of sudden or fast movement. Here, record as time for one walking step is 1 s and step
length is 20 cm. Kinematic diagram of dynamic walking is given in Fig. 7. When the required
peak error in ZMP value is very small, usually simulation time will increase or at times it may
not give optimum solution. During attempts with increased number of iterations it was found
that simulation takes more time and peak error value obtained is higher than in the static case.
Hence a reasonable limiting value of Peak error in ZMP is selected, (XzmpError, 0, 0) ≤ (1, 0, 0).
Dynamic stability is achieved by changing the position of the hip and upper body of the
biped. Stability of the biped is influenced more compared to other links because the angular
positional variations of fifth, sixth and seventh links as shown in Fig. 8. Various plots given
below are for the first set of simulation from the dynamically stable simulation results given
in Table III. It is clear from the table that the time taken for all simulations is more or less the
same. Hip positions are the same and reasonably good in the 2nd
, 3rd
and 4th
simulation but
minimum peak error is obtained in 5th
and 6th
simulation.
Angular velocity of the fifth link at the upper body is more as given in Fig. 9.This depicts
the influence of the fifth link in stable walking. Angular acceleration, linear velocity and
linear accelerations of lower leg of the swing foot as shown in Figs. 10, 11 and 12 are of high
value because the right foot has to move in a cycloidal path for the step length of 20 cm in 1 s
and also this lower leg has greater significance in the minimization of peak error in ZMP. It is
clear from the Fig. 13 that the fitness value is much higher up to around 400 iterations and
then reached the required minimum peak error value of 0.9996 after the 641 iterations. The
simulation took a reasonable time of 268.56 seconds. Up and down in the fitness plot
symbolizes the ability of simulated annealing algorithm to achieve global optimum by
escaping from the local minimum because it is independent of gradient descent and initial
conditions. Peak error in ZMP is plotted in Fig. 14. ZMP plot gives the idea of the foot size
variation in X direction for the optimum stable gait generation.
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
187
Figure 7: Dynamic walking.
Figure 8: Variations of joint angles.
-250
-200
-150
-100
-50
0
50
100
150
200
250
300
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
An
gu
lar
Accele
rati
on
(ra
d/s
2)
Ang acc of stance
leg-link 2
Ang acc of stance
leg-link 3
Ang acc of upper
body-link 4
Ang acc of upper
body- link 5
Ang acc of swing
leg-link 6
Ang acc of swing
leg-link 7
Figure 9: Variations of angular velocity. Figure 10: Variations of angular acceleration.
Join
t A
ng
les
(Deg
rees
)
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
188
Figure 11: Variations of linear velocity. Figure 12: Variations of linear acceleration.
Figure 13: Variations of fitness function. Figure 14: Variations of ZMP.
Table III: Comparison of parameters in dynamically stable simulation.
Simulation
No.
Minimum peak
error (< 1) [cm]
Time
taken [s]
No. of
iterations
Variables
hs = [rx, ry, xhi, yhi, θ4i, θ4f, θ5i, θ5f ]
1 0.9996 268.56 641 [ 1.2849 2.5492 4.2741 62.5425
0.4807 0.2091 0.2016 0.6929 ]
2 0.9998 260.41 638 [ 0.3459 1.9946 -3.5998 62.2865
0.2492 0.1094 0.6027 0.1156 ]
3 0.9998 266.39 644 [ 0.3459 1.9946 -3.5998 62.2865
0.2492 0.1094 0.6027 0.1156 ]
4 0.9998 262.47 638 [ 0.3459 1.9946 -3.5998 62.2865
0.2492 0.1094 0.6027 0.1156 ]
5 0.9735 246.98 574 [ 0.3778 2.1674 -4.2163 62.0183
0.2599 0.0408 0.6642 0.5454 ]
6 0.9735 241.47 568 [ 0.3778 2.1674 -4.2163 62.0183
0.2599 0.0408 0.6642 0.5454 ]
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
189
7. CONCLUSIONS AND OUTLOOK
Stable optimum gait generation of 8 links biped robots is presented in this work. Simulation
clearly illustrates the influence of hip and upper body positions for stable walking with the
minimum foot size. From the literature it is noted that, in most of the planar robot walking
cases, upper body is not considered and this is the only work carried out with two links in the
upper body. This paper demonstrated that the proposed gait planning method can achieve
statically and dynamically stable biped gait with small values of ZMP. When the foot size of
the biped is large, it is very difficult to move easily because of external and self collision. In
application point of view, design and control of smaller foot is better even though the stable
locomotion is difficult. Currently, researchers are trying to minimize the foot size for getting
easy, fast and stable walking. This work helps to decide the foot dimensions. During walking,
various motion parameters such as joint angles, velocity, accelerations, linear velocities and
accelerations are plotted and analyzed. Proposed SA algorithm gives better solution in a
reasonable time. The variation of fitness function shown in Figs. 6 and 14 indicate the
significance of SA in the optimization problems. The use of simulated annealing optimization
algorithm is not reported by other researchers specifically in the area of humanoid or biped
robotics. These type of work helps to design joints and to select the proper actuators by
realizing the utility of various joints in the optimum stable walking. It is seen that this
algorithm can be used for the optimization of various gaits for stair ascending and descending,
stepping over obstacle, ditch crossing and obstacle avoidance etc. The analysis given above is
restricted to the motion in sagittal plane. However it is possible to extend this methodology
for motion in the frontal plane also.
REFERENCES
[1] Huang, Q.; Yokoi, K.; Kajita, S.; Kaneko, K. (2001). Planning Walking Patterns for a Biped
Robot, IEEE transactions on Robotics and Automation, Vol. 17, No. 3, 280-289,
doi:10.1109/70.938385
[2] Vukobratovic, M.; Andric, D.; Borovac, B. (2004). How to achieve various gait patterns from
single nominal, International Journal of Advanced Robotic Systems, Vol. 1, No. 2, 99-108
[3] Dip, G.; Prahlad, V.; Kien, P. D. (2009). Genetic algorithm based optimal bipedal walking gait
synthesis considering tradeoff between stability margin and speed, Robotica, Vol. 27, No. 1, 355-
365, doi:10.1017/S026357470800475X
[4] Shrivastava, M.; Dutta, A.; Saxena, A. (2007). Trajectory generation using GA for 8 DOF biped
robot with deformation at the sole of the foot, Journal of Intelligent Robotic Systems, Vol. 49,
No. 1, 67-84, doi:10.1007/s10846-007-9129-x
[5] Juang, J.-G. (2001). Intelligent trajectory control using recurrent averaging learning, Journal of
Applied Artificial Intelligence, Vol. 15, No. 3, 277-296, doi:10.1080/08839510151063253
[6] Fu, C. L.; Chen, K. (2008). Gait synthesis and sensory control of stair climbing for a humanoid
robot, IEEE Transactions and Industrial Electronics, Vol. 55, No. 5, 345-358
[7] Goncalves, J. B.; Zampeieri, D. E. (2003). Recurrent neural network approaches for biped
walking robot based on zero moment point criterion, Journal of the Brazilian society of
Mechanical Sciences and Engineering, Vol. 25, No. 1, 1678-1707, doi:10.1590/S1678-
58782003000100010
[8] Ito, S.; Amano, S.; Sasaki, M.; Kulvanit, P. (2008). A ZMP feedback control for biped balance
and its application to in-place lateral stepping motion, Journal of Computers, Vol. 3, No. 8, 23-
31, doi:10.4304/jcp.3.8.23-31
[9] Bigdeli, N.; Afsar, K.; Lame, B. I.; Zohrabi, A. (2008). Modeling of a five link biped robot
dynamics using neural networks, Journal of Applied Science, Vol. 20, No. 1, 3612-3620
Sudheer, Vijayakumar, Mohandas: Optimum Stable Gait Planning for an 8 Link Biped …
190
[10] Park, J. H.; Choi, M. (2004). Generation of an optimum gait trajectory for biped robot using a
genetic algorithm, JSME International Journal, Vol. 47, No. 2, 715-721,
doi:10.1299/jsmec.47.715
[11] Ja, R. K.; Singh, B.; Pratihar, D. K. (2005). On-line stable gait generation of a two legged robot
using Genetic – Fuzzy system, Journal of Robotics and Automation Systems, Vol. 53, No. 1, 15-
35
[12] Vundavilli, P. R.; Pratihar, D. K. (2009). Soft computing based gait planners for a dynamically
balanced biped robot negotiating sloping surfaces, Journal of Applied Soft Computing, Vol. 9,
No. 2, 191-208, doi:10.1016/j.asoc.2008.04.004
[13] Kirkpatrick, S.; Gelatt, C. D.; Vecchi, M. P. (1983). Optimization by Simulated Annealing,
Science, New Series, Vol. 220, No. 1, 671-680