Dynamic Control of Legged Humanoid Robots

This Helmholtz Young Investigators Group project, titled “Dynamic Control of Legged Humanoid Robots”, covers key aspects of humanoid robot control by focusing on advanced motion skills for locomotion and physical interaction.

During the last 20 years, control applications in robotics research have been extended from the control of fully actuated serial- and closed-chain manipulator systems to larger systems of increased complexity including mechanical systems with non-holonomic constraints, under-actuation in the control input, and varying contact constraints leading to hybrid dynamical systems. Today, biped humanoid robots serve as an important application field of robot control in which all the above mentioned control problems appear at the same time.

While basic walking control solutions for bipedal robots have been developed based on simplified models, these approaches still lack in robustness with respect to the environment model, e.g. the foot ground contact, and do not allow a generalization to the full dynamical model. Robust bipedal walking is a key element if one seriously considers employing humanoid robots as our mechanical servants in domestic living environments. In this project, walking control algorithms will be developed, which allow for a real-time step adaptation and thus an online adaptation to the environmental conditions.

Compliant whole body motion control is a must both for physical interaction with humans and for autonomous manipulation. In particular, when considering robots within the direct surroundings of humans, the problem of handling intentional and unintentional physical interaction with humans during standing as well as during locomotion has to be mastered. While state of the art humanoid robots are able to fulfil a large range of controlled motions in well defined environments, nowadays locomotion control approaches focus on static environments and consider no or only minimal physical interaction with humans. In this project, we aim not only at reactive skills to physical contact enforced by a human, but moreover also consider active physical human robot interaction initiated and performed by the robot.

Walking skills between robots and humans differ not only in the control, but also in the physical basis. Biomechanics research has found the existence of stable walking motions based on simple, but human oriented, compliant models. While these works focus on basic principles of bipedal locomotion, the utilization of these principles for robot control is a still widely open topic. This project aims at transferring anthropomorphic walking principles to robotic systems in order to make considerable progress in the performance of robotic walking. This expected increase in performance can be evaluated in quantitative terms like speed or energy consumption, as well as in the qualitative achievement of advanced locomotion modes like running or jumping. The utilization of anthropomorphic walking principles will involve the design of new hardware platforms as well as control oriented research within the Helmholtz Young Investigators Research Group.

A more explicit skill transfer from the human to robots becomes interesting for highly articulated humanoid robots with a large number of degrees-of-freedom. Here, teaching by demonstration provides a means to generate human-like motion behaviour, without being too sensitive to specific optimization criteria. Within this project, learning of human motion skills will focus on locomotion and force based manipulation skills. Since biped balancing and locomotion is an inherently unstable process, this requires an extension from pure imitation of the human’s motions to an ‘emulation’ of the locomotion skill.