Manual Control of Bipedal Walking

To compare with "Reinforcement Learning for CPG-controlled Bipedal Walking",
let us try to control the identical model of bipedal walking manually.
You might understand its difficulty.

[How to control]

By double-clicking the blue or purple horizontal scroll bars, an initial horizontal velocity is given to the model (only once).
By moving the blue (right foot) or purple (left foot) horizontal scroll bars, you can change the destination angles of both feet.
By changing the destination angle of one foot, that of the other foot is determined automatically.

If the above application does not start, please install Java from here. >>

The model of biped walking (Taga, 1991) is written by an equation of motion of 14 variables

with 8 constraints.

The position of the model is indicated by 6 variables because 14 - 8 = 6, and they are the position (x, y) of the hip, and the angles of hips and knees (θ_R1, θ_R2, θ_L1, θ_L2) for both feet.

Moreover, by considering their time-derivatives (v_x, v_y) and (ω_R1, ω_R2, ω_L1, ω_L2), the state of this dynamical system can be described.

The torque T given to this model is determined by PD control scheme.

In this applet, the destination angles for both feet θ_R1^d and θ_L1^d can be changed with two horizontal scroll bars.

θ_R2^d and θ_L2^d are determined automatically so that θ_R2^d=55 when θ_R1^d > 0 and ω_R1^d > 0 are satisfied, and otherwise θ_R2^d=0.

This page is based on the following paper.

G. Taga, Y. Yamaguchi, and H. Shimizu,
"Self-organized control of bipedal locomotion by neural oscillators in unpredictable environment"
Biological Cybernetics, vol. 65, pp.147-159 (1991).

<< Reinforcement Learning for CPG-controlled Bipedal Walking / Cart-Pole Balancing by Reinforcement Learning >>

Back to Takashi Kanamaru's Web Page