Humanoid Robot Pitch Axis Stabilization using Linear Quadratic Regulator with Fuzzy Logic and Capture Point

TL;DR

This paper presents a hybrid control approach combining LQR, fuzzy logic, and system identification to stabilize a humanoid robot's pitch axis during standing and walking on challenging surfaces.

Contribution

It introduces a novel control system integrating LQR with fuzzy logic and system identification for improved humanoid robot stability.

Findings

The control system effectively stabilizes the robot against disturbances.

Fuzzy logic enhances LQR performance under nonlinear conditions.

System identification accurately models robot dynamics for control design.

Abstract

This paper aims for a controller that can stabilize a position-controlled humanoid robot when standing still or walking on synthetic grass even when subjected to external disturbances. Two types of controllers are designed and implemented: ankle strategy and stepping strategy. The robot's joints consist of position-controlled servos which can be complicated to model analytically due to nonlinearities and non-measurable parameters, hence the dynamic model of the humanoid robot is acquired using a non-recursive least squares system identification. This model is also used to design a Kalman Filter to estimate the system states from noisy inertial measurement unit (IMU) sensor and design a linear quadratic regulator (LQR) controller. To handle the nonlinearities, the LQR controller is extended with fuzzy logic algorithm that changes the LQR gain value based on angle and angular velocity membership functions. The proposed control system can maintain the humanoid robot's stability around the pitch axis when subject to pendulum disturbances or even restraining force from a spring balance.