Site Search  

Resources » Publication Details

Publication Details

Reference TypeConference Proceedings
Author(s)Rotella, N.; Herzog, A.; Schaal, S.; Righetti, L.
TitleMomentum Planning, Estimation and Control for Force-Centric Bipedal Locomotion
Journal/Conference/Book TitleDynamic Walking 2015 (Talk, Poster)
Keywordshumanoid robot, model predictive control, state estimation, force control
AbstractThe stability of a humanoid robot is ensured by constraining its Center of Pressure (CoP) within its support polygon; us- ing a simplified model such as the Linear Inverted Pendulum Model (LIPM) allows for a linear mapping between Center of Mass (CoM) and CoP motion. This enables the use of linear Model Predictive Control (MPC) methods to generate a CoM trajectory which realizes a desired CoP motion for the LIPM [1]. Traditional approaches to walking resolve the planned motion for joint trajectories and track these using stiff position control. However, the CoP is defined by interaction forces - not positions. Since position control offers no direct control of force, this method relies on trajectory engineering and tun- ing. On a torque-controlled robot, however, one can directly track LIPM-consistent forces rather than COM motion. This allows for control of impedance, yet even this fails to work well without tuning for several reasons. First, the simplified model forces - even if reproduced exactly on the robot - will not generate the desired CoP motion due to model differences between the robot and the LIPM. Second, inverse dynamics solvers do not account for discrepancies between planned and measured forces due to perturbations, unobserved terrain and so on. In order to generate dynamic reactive behaviors on the real robot, we need to use descriptive simplified models which are consistent with the full dynamics and make use of endeffector force/torque (F/T) sensors for state estimation.
Link to PDF

Designed by: Nerses Ohanyan & Jan Peters
Page last modified on June 20, 2013, at 07:00 PM