The 30th Annual International Conference of Iranian Society of Mechanical Engineers

10 to 12 May, 2022, Tehran, Iran.

ISME2022-IC1332

Optimal Motion Generation of the Bipedal Under-Actuated Planar Robot for Stair Climbing
Aref Amiri 1, Hassan Salarieh 2

1
Graduate Student, Sharif University of Technology, Tehran; [email protected]
2
Professor, Sharif University of Technology, Tehran; [email protected]

Abstract using data based on motion capturing and then try to

The importance of humanoid robots in today's world is optimize their results by combining optimization
undeniable, one of the most important features of methods [8]. However, according to the existing
humanoid robots is the ability to maneuver in literature, few articles have attempted to design a
environments such as stairs that other robots can not holonomic path for under-actuated bipedal robots
easily cross. A suitable algorithm to generate the path without feet. Due to the importance of optimal motion
for the bipedal robot to climb is very important. In this planning, a lot of work has been done in recent years in
paper, an optimization-based method to generate an this area.
optimal stairway for under-actuated bipedal robots In this paper, the problem of motion planning is
without an ankle actuator is presented. The generated investigated to find the optimal paths for under-actuated
paths are based on zero and non-zero dynamics of the bipedal robots to step on the stairs, the results obtained
problem, and according to the satisfaction of the zero as a control output will cause the robot to move properly
dynamics constraint in the problem, tracking the path is and optimally. This article consists of three sections. In
possible, in other words, the problem can be the first part, the dynamic model of the bipedal robot is
dynamically feasible. The optimization method used in derived. In the second part, the constraints of the
the problem is a gradient-based method that has a optimization problem are examined, in the third part, the
suitable number of function evaluations for cost function and method of optimal problem solving
computational processing. This method can also be and finding a suitable movement gate are examined. In
utilized to go down the stairs. the fourth section, the results are presented and
discussed, and at the end, the research of this article is
Keywords: Bipedal robot, under-actuated, optimization, summarized as the conclusion.
motion planning
Dynamics equation
Introduction The dynamic model of the robot is shown in Figure
Inspired by human body physics, bipedal robots have 1. The robot has 7 degrees of freedom and 5 links,
many degrees of freedom and can perform various each leg has two joints (one in the knee and the
actions with their joint movements. Bipedal robots can other in the hip) and 3 degrees of freedom. We
adapt to different environments that other wheeled assume that the contact of the tip of the leg is the
robots are unable to move. The study of path (trajectory) point.
generation methods as a reference for the output of the
control problem of bipedal robots in this regard is
essential. For the bipedal robot to climb the stairs, it is
necessary to analyze the movement of them ascending
the stairs and to examine the method of planning the
bipedal robot to move and to determine the position of
feet for walking on the stairs [1].
So far, researches have been done on how to go up
and downstairs and find a suitable or optimal path for
bipedal robots. Various papers using optimization
algorithms and considering the robot angles as
polynomial functions tried to design an optimal path for
a 6-degree bipedal robot [2]. Some articles have even
paths planned for multi-legged robots to cross the stairs
[3]. Some articles also used stability criteria such as
ZMP in designing their paths [4-7]. But this method is
only appliable for robots that have feet (soles) with ankle
joint actuators, which often have much lower speed in
maneuvering than under-actuated robots without feet,
Figure 1. Planar bipedal robot
and of course, due to the relatively large feet have more
wasted energy. Some articles also derive their initial path

10 to 12 May, 2022
 The robot's motion is planar and the robot has 4 Coefficient of friction limit: for the robot to move on
actuators, two actuators at the knees and two actuators real environments, the ratio of horizontal force to
at the junction of the hip and the trunk so that there is vertical force should not be more or less than a certain
one actuator between each leg and trunk. It is assumed limit. In other words, the coefficient of friction required
that by hitting the tip of the swing leg on the ground, the for stepping should not exceed a certain limit that can
other leg rises from the ground, in other words, the not be implemented in real environments. (inequality
robot has no double support phase. So, when moving on constraint).
the stairs, no time is wasted for placing both feet on the 2. Kinematic constraints:
ground. Therefore, the hybrid dynamic equations of a Configuration: As an initial and final condition, the
robot are a combination of a single support phase and r1 robot needs to move from an initial configuration to a
r2
collision phase. The equations of the hybrid model are final configuration. The best option is for the initial and
as follows: final state to be the same so that the robot has
 x  f ( x)  g ( x)u x   (1) periodicity in its movement and the best footprint is in
:   the middle of each stair Figure 2 (equality constraint).
 x  ( x ) x  
The vector x : (qT , qT )T consists of the vector of
generalized coordinates and their derivatives.  is a map best
footprint
to find the states of the system exactly after the collision,
and the positive and negative symbols indicate the states clearance height
of the system before and after the collision. The switch
condition is as follows: width

  (q, q)  x | P2v (q)  0, P2h (q)  0 (2) Figure 2. Stair properties

In equation (2), P2h represents the horizontal position Angular velocity limit: Because motors have limited
v
angular velocity production. (inequality constraint)
of the swing leg and P represents its vertical position.
2 Contact in single support phase: The robot is in
contact with the ground during the single support phase
The dynamic equations of the robot before and after and the acceleration of the contact point in the
the collision and in the single support phase can be horizontal and vertical direction during this period is
written as follows: zero. (equality constraint)
M ( q ) q  C( q ,q ) q  G q   B( q )u (3) Swing leg collision: The robot swing leg during the
Matrix B is also a pre-multiplication matrix in the single-phase phase, except at the beginning and end of
torque vector and is not a square matrix due to the the phase, should not collide with the ground, on the
under-actuation of the system. other hand, should have a suitable distance to the
In Equation 1, there is an expression called zero obstacles.
dynamics, and it is easy to separate this term if the Knees movement limitation: To create maximum
generalized coordinates of the system are written in similarity to human movement, the robot knee should
relative terms (as has been done in this paper). The not be opened and closed too much.
satisfaction of this constraint is important in two ways. Failure to satisfy any of the above constraints will cause
First, if this constraint is not satisfied, the problem of problems in creating optimal and appropriate
optimizing the input torques is practically ambiguous, movement.
because these torques are not really applicable to the
problem. Although it may lead to a feasible kinematic Optimization method
equation (kinematically possible), it is not feasible in This optimization is a nonlinear, constrained, and single-
terms of control (open-loop), i.e. it is not dynamically objective problem.
possible. Cost function: To find the optimal path, various cost
functions are considered, for example, the norm of
Optimization problem torque input, system input energy, and cost of transport
The most important constraint of the problem, called are common options. In this paper, we consider the
zero dynamics, was introduced in the previous section. norm of torque inputs as the cost function. By this
Other constraints in this issue are important to plan the choice, the torques are rational in size and will have
robot movement in the best way; the constraints of the proper distribution (If the optimization problem is
optimization problem are generally classified into two solved properly).
4
J   ( ui2 ( )) d
general modes of constraints based on dynamics and T
(4)
constraints based on kinematics. 0
i 0
In the above equation, T is the length of the time
1. Dynamic constraints: period.
Torque limit: because the torque generators have a Selection of optimization variables: Optimization
certain limit (inequality constraint). variables can have different types, one of the best
Zero dynamic: the importance of which was choices is the paths followed by generalized
mentioned earlier (equality constraint). coordinates. Here our choice is a time-varying path as a
function of polynomials. The polynomial functions are

10 to 12 May, 2022
uniform and smooth, and they are also simple for
deriving.
n4
qk (t )    k ,i t i (5)
i 0
The degree of this polynomial must be chosen in
such a way that the number of optimization parameters,
which are the same as the number of polynomial
coefficients, are appropriate (minimum value to have a
smooth motion satisfied the mentioned constraints). In
this article, we choose the function of order 4 to have
freedom of action in terms of the optimization problem
and also not to make the number of optimization
parameters of the problem irrational and complicated.
Method of solving the optimization problem: This
optimization problem is solved by Variable Metric
Figure 3. Input torques
methods for constrained optimization. This method is a
gradient-based method, which provides a desirable and
fast solution. Another advantage of this method is to not According to Figure 3, the torques have a good
get out easily from the feasible area [9]. margin from the saturation and compared to other
articles and research reviewed in the introduction, more
Results and Discussion optimal results have been obtained, also zero dynamics
Following the model and algorithm presented above, a (  v ) in a very good way is satisfied.
bipedal robot has been simulated to climb the stairs. The
height of the stairs is considered 20cm and the width of
the stairs is 40cm. The robot model specifications are in
accordance with Table 1. The initial and final angles of
the bipedal robot as a configuration are given in Table
2. Here the initial and final configurations are intuitively
obtained from the human configuration. The speed of
crossing each step is .5 seconds. The torque limit
applied to the system is 150 N.m and the maximum
angular velocity of the motors 10 rad/sec can be.

Table 1. Rabbit robot properties [10]

Symbol Value
m1, m5 3.2 kg
m2, m4 6.8 kg
m3 20 kg Figure 4. Friction coefficient
I1, I5 0.93 kg-m2
I2, I4 1.08 kg-m2
I3 2.22 kg-m2 According to Figure 4, it is clear that the generated
l1, l5 0.4 m path needs the maximum coefficient of friction .69 to
l2, l4 0.4 m slip, so on all surfaces that have a coefficient of friction
l3 0.625 m
d1, d5 0.128 m higher than .69 there is the ability to move.
d2, d4 0.163 m
d3 0.2 m

Table 2. The initial and final configuration

Parameters Initial value(rad) Final value(rad)
q1 0.2618 0.1964
q2 1.3140 0
q3 -1.2267 0.0219
q4 -0.0219 1.2267
q5 0 1.3140

Figure 5. Angles vs. angular velocities

According to Figure 5, the generated paths, due to

the nature of the polynomial functions, have a smooth

10 to 12 May, 2022
and non-breaking behavior, and the angular velocities 2021 IEEE International Conference on Robotics
are far from their saturation limit. and Automation (ICRA). IEEE, 2021.
[4] Kim, Eun-Su, Jo-Hwan Kim, and Jong-Wook Kim.
"Generation of optimal trajectories for ascending
and descending a stair of a humanoid based on
uDEAS." 2009 IEEE International Conference on
Fuzzy Systems. IEEE, 2009.
[5] Sugahara, Yusuke, et al. "Walking up and down
stairs carrying a human by a biped locomotor with
parallel mechanism." 2005 IEEE/RSJ International
Conference on Intelligent Robots and Systems.
IEEE, 2005.
[6] Zhang, Qin, et al. "Action generation of a biped
robot climbing stairs." 2013 IEEE International
Conference on Mechatronics and Automation.
IEEE, 2013.
[7] Kim, E., T. Kim, and J-W. Kim. "Three-dimensional
modelling of a humanoid in three planes and a
motion scheme of biped turning in standing." IET
Figure 6. Stick diagram of the climbing a stair up control theory & applications 3.9 (2009): 1155-
1166.
As can be seen in Figure 6, the robot's movement is [8] Powell, Matthew J., Huihua Zhao, and Aaron D.
quite normal and very similar to human movement. The Ames. "Motion primitives for human-inspired
trunk is kept in a good position and also the tip of the bipedal robotic locomotion: walking and stair
feet and other links do not touch the surfaces except at climbing." 2012 IEEE International Conference on
the beginning and at the end of the movement. Robotics and Automation. IEEE, 2012.Misc, A.,
According to the sum of the presented results, the 2003. Miscellaneous Title. On the WWW, May.
generated path is an optimal path for the proper gait of URL http://www.abc.edu.
the under-actuated bipedal robot. [9] Powell, Michael JD. "A fast algorithm for
nonlinearly constrained optimization calculations."
Conclusions Numerical analysis. Springer, Berlin, Heidelberg,
In this article, we present a method to generate optimal 1978. 144-157..
motion for a bipedal robot, we used this method to find [10] Chevallereau, Christine, et al. "Rabbit: A testbed for
the paths that the 'rabbit' robot by tracking them can advanced control theory." IEEE Control Systems
optimally climb stairs. This process consists of 3 parts: Magazine 23.5 (2003): 57-79.
robot dynamic extraction (because optimization is based
on the model), design of constraints based on dynamics
and kinematics, and optimization. As a result of the
problem, a series of virtual holonomic paths were
extracted in which the zero hybrid dynamics of the
problem is also satisfied, so tracking the paths are
possible for under-actuated robots.
In the future, we plan to use a new method called impact
invariance to design the above path, which guarantees
the periodicity of the proposed paths.

References

[1] Goldfarb, Nathaniel, Charles Bales, and Gregory S.

Fischer. "Toward Generalization of Bipedal Gait
Cycle During Stair Climbing Using Learning From
Demonstration." IEEE Transactions on Medical
Robotics and Bionics 3.2 (2021): 446-454.
[2] Kweon Soo Jeon, Ohung Kwon, Jong Hyeon Park.
Optimal trajectory generation for a biped robot
walking a staircase based on genetic
algorithms[C]//Proceedings of 2004 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, Sendai, Japan: IEEE, 2004: 2837- 2842.
[3] Cebe, Oguzhan, et al. "Online dynamic trajectory
optimization and control for a quadruped robot."

10 to 12 May, 2022