Stable Modifiable Walking Pattern Algorithm with Constrained Optimized Central Pattern Generator

Abstract

Abstract. In this paper, stable modifiable walking pattern algorithm is proposed using evolutionary optimized central pattern generator (CPG). Sensory feedback pathways in CPG are proposed, which use force sensing resistor (FSR) signals. For the optimization of CPG parameters, two-phase evolutionary programming (TPEP) is employed. Modifiable walking pattern generator (MWPG) generates position trajectory of center of mass (COM) of humanoid robot and CPG generates sagittal swing foot position trajectory. The effectiveness of the proposed scheme is demonstrated by simulations using a Webots dynamic simulator for a small sized humanoid robot, HSR-IX, developed in the Robot Intelligence Technology (RIT) Lab, KAIST.

Mechanism, Actuation, Perception, and Control of Highly Dynamic Multilegged Robots: A Review

“…The CPG controller can not only produce well-coordinated leg movements but also accomplish gait transitions with simple descending control signals [145]. Moreover, the controller can use a neural oscillator (NO) with reinforcement learning to obtain parameters online [146] and optimize various parameters [147]. A great deal of previous research regarding the CPG model attempted to generate dynamic locomotion, such as in the Tekken 1 and 2 [105][106][107], Cheetah-cub [40], HyQ [17], and Baby elephant [108].…”

“…There are two obvious trends for CPG control: (i) the external sensory information, such as the robot torso's postures [105], touchdown feedback [107], and leg loading [148], is applied to the CPG model to improve control performance; (ii) the CPG is combined with another control method, such as inverse dynamics [17,108], RL [143,146,147], or VMC [149]. There are also some issues that need to be further studied.…”

Versatile and robust 3D walking with a simulated humanoid robot (Atlas): A model predictive control approach

“…Model-free methods like [5], [6] and [7] optimize a walking pattern off-line, targeting stability, fast motion, energy efficiency or similarity to human walking. Other approaches incorporate a kinematics model of the robot into the loop, which enables them to convert Cartesian positions, velocities or forces to joint variables and vice-versa.…”