Session: MR-5-2: Motion Planning, Dynamics, and Control of Robots

Paper Number: 116945

116945 - Control of a Bipedal Walking Using Partial Feedback Linearization and Gaussian Process Regression-Based of the Step-to-Step Map 

Bipedal robots with small and point feet find it  difficult to balance instantaneously due to under-actuation, but can be relatively easy to control over the time scale of a step. Thus, models of step-to-step dynamics provide an effective means for the development of step-to-step controllers. Such models are analytically intractable because they are obtained from integrating nonlinear equations of motion. Our hypothesis is that obtaining approximations of such models can be effective for real-time computations of control inputs. We use partial feedback linearization to reduce the system dimensions for the step-to-step map. The system is reduced to one uncontrolled degree of freedom at the step-to-step map. With a data-driven approach we use Gaussian Process Regression to approximate the step-to-step map and its region of validity. We formulate and solve an optimal control problem with the approximations to compute control for tracking a reference velocity and walking on stepping stones. These solutions demonstrate the need for bipedal robots to achieve specific control objectives such as walking speed and foot placement in order to navigate through restricted environments such as the stepping stones. We provide simulations on a 5-link biped to show the efficacy of the approach. A video link is provided: https: //www.youtube.com/watch?v=p2TLU3Jzul8.

Presenting Author: Daniel Torres University of Illinois at Chicago

Presenting Author Biography: A 3rd year PhD student at the University of Illinois at Chicago.

Authors:

Daniel Torres University of Illinois at Chicago
Ernesto Hernandez Hinojosa University of Illinois at Chicago
Pranav Bhounsule University of Illinois at Chicago