ResearchDigital Human Modeling Research Group

Balancing Control of Biped Walking by Memory Based Control Method

>> Japanese

Improvisational Footwork: Bipedal Walking Control Based on Shifting of Multiple Trajectories

For the Walking with High Tolerance to the Environmental Changes

Bipedal walking is expected to have the potential to overcome various kinds of rough terrain.
The examples of the rough terrain are a carpet, mattress, gravel road, grass field, scrag, mud, snow, etc. They have difficulties to predict the contact state and of motion planning based on the environmental modeling.
To cope with the difficulties, (1) the balancing controller needs to robust to the contact state fluctuation, (2) the balancing controller needs to adapt immediately to the changing states.

Balancing with free of influence from foot contact
The balancing controller, which is not greatly affected in the contact state, is necessary to use a basic character to approve regardless of the state of the contact.
To abstract the element, and to decrease the relating factor, the model was simplified to the minimum requirement. The foot-ground contact is abstracted to the point contact (pivot), and the whole body is considered to be a compound pendulum of the inverted pendulum (stance leg) and the common pendulum (swinging leg). The two dimensional motion is restricted to the sagittal plane.


Compass-like Bipedal Walking Model


For this model, the occurrence of the fall is consolidated in two patterns.
  1. Stumble = foot doesn't arrive at stumbling body -> fall
  2. Upset = the tip of foot goes on except body -> fall
So as not to fall, the relation of position and the speed between the upper body (whole body center of gravity) and feet (contact point) should be kept appropriate.
Immediate Motion Generation and Control
The technique of the memory based motion control named the Multiple Trajectory Shifting Control is used.
It steps on two stages in the control. (1) Making of the memory consists of well walking motion patterns, (2) Walking motion generation and control by referring the memory.

(1) Preparing a deliberate trajectory library by various paces and strides, beforehand.


Stage of memory generation in a variety of walking patterns


The parameter set of the pace and the stride (movement target value of articulatio coxae), and the stance leg trajectory (approximates the state transition between the whole body center of gravity and the contact point) is associated, and preserved.

(2) The set of the measured stance leg states is made a key, and the motion generation is done while referring.


Stage of walking control by memory reference


The set of the stance leg states at the moment is made a key every sample time, and the set of the parameter of corresponding pace and stride is drawn out from the library. The parameter set is addressed to the desired value of articulatio coxae, and the swinging leg is controlled (The stepping motion in walk is reproduced). The motion generation and control are expected to accomplish because it is reproduction of the memory which consists of the successful walking motions.
Simulation result

Regular walking. The desired hip joint angle (equivalent to the step length) is 0.6[rad].


Regular walking.



It keeps changing the carrying of the foot in compliance with perturbations, and the walking balance still being maintained.


Big perturbation (80[kg]) is added while moving to the right direction (left figure), and stepping back after the impact (right figure).


Walking on horizontal quaking ground (frequency 3[Hz] and amplitude 5[cm]).


Walking on unleveled ground. The ground is generated with the line segments by random values. The amplitude of the control point at each segment end is 5cm (10cm in the maximum swinging width).


Walking, when the random seed for generating the unleveled ground is ten.


Walking, when random seed is 1, 2, 3, 4, 5, 6, 7, 8, 9 .



Up-hill(0.2[rad]) walking.


Down-hill(0.3[rad]) walking.



Walking by a variety of body parameters. The standard body is 40[kg] in the mass of the upper part, and 1[m] in the natural length of the leg.


Walking, when the mass of upper part of the body is increased up to 110[kg].


Walking, when the mass of upper part of the body is decreased down to 20[kg].


Walking, when the natural length of the leg is extended up to 1.65[m].


Walking, when the natural length of the leg is shortened down to 0.75[m].


The leg structure (the knee joints) is changed to a rotational type from the prismatic type, while the controller is the same.


Walking with kneed legs.

The structure of the upper part of the body is changed from the sphere shape (point mass) to the rectangular box shape (distributed mass), while the controller is the same.


Walking with the rectangular box shape upper body.


Walking by a variety of sole shape. The walking motion pattern is heel-toe. Strength of the push off motion is adjusted in correspond to each sole shape. The figures are aligned in order that the push-off motion can be small.


Walking with a round shape sole.


Walking with a wooden clogs shape sole.


Walking with a shoes shape sole.


Walking with a flat shape sole.



Obstacle avoidance by temporary change of the desired step length. The standard step length is 0.6[rad] as the above-mentioned.


Walking, when the desired step length is temporarily expanded up to 0.9[rad].


Walking, when the desired step length is temporarily narrowed down to 0.3[rad].

Alteration of walking pace by change in the regular desired step length (The amount of push-off is also temporarily adjusted). The standard desired step length is 0.6[rad], meanwhile the narrow one is 0.4[rad], and the wide one is 0.8[rad].


Walking, when expanding the step length after it is narrowed.


Walking, when narrowing the step length after it is expanded.



Under Study. Recovery motion on the slippy area. (slip and fall on ice.)



The biped slips and falls down on the area where the friction coefficient is very small.