Using efference copy and a forward internal model for adaptive biped walking

Abstract

To behave properly in an unknown environment, animals or robots must distinguish external from selfgenerated stimuli on their sensors. The biologically inspired concepts of efference copy and internal model have been successfully applied to a number of robot control problems. Here we present an application of this for our dynamic walking robot RunBot. We use efference copies of the motor commands with a simple forward internal model to predict the expected self-generated acceleration during walking. The difference to the actually measured acceleration is then used to stabilize the walking on terrains with changing slopes through its upper body component controller. As a consequence, the controller drives the upper body component (UBC) to lean forwards/backwards as soon as an error occurs resulting in dynamical stable walking. We have evaluated the performance of the system on four different track configurations. Furthermore we believe that the experimental studies pursued here will sharpen our understanding of how the efference copies influence dynamic locomotion Electronic supplementary material The online version of this article (doi:10.1007/s10514-010-9199-7) contains supplementary material, which is available to authorized users.J. Schröder-Schetelig Max Planck Institute for Dynamics and Self-Organization, Bunsenstraße 10, 37073 Göttingen, Germany e-mail: johannes.schroeder-schetelig@ds.mpg.de P. Manoonpong · F. Wörgötter ( ) Bernstein Center for Computational Neuroscience (BCCN), University of Göttingen, Friedrich-Hund-Platz 1, 37077 Göttingen, Germany e-mail: worgott@physik3.gwdg.de P. Manoonpong e-mail: poramate@physik3.gwdg.de control to the benefit of modern neural control strategies in robots.

Adaptive Quadruped Locomotion: Learning to Detect and Avoid an Obstacle

“…The value held by the STM cell is expected to last until a learning signal is triggered. The expected behavior to the STM cell is achieved through equation 3 28. and its displayed in figure 5.20.There, despite the initial convergence error, all the activations have its value decayed in time, as expected.…”

“…In [28], the authors used the biologically inspired concepts of FIM and efference copy to detect changes in the ground's slope. The detected changes are then used to stabilized a biped robot locomotion -which movements are restricted to the saggital plane -when the ground's slope changes.…”

“…Also, the responses in the upper body component are predefined and designed offline. Although the type of approach in [28] uses the same concepts as ours, their solution does not address the online learning of the internal model or the actuator's responses. Furthermore, our solution keeps the learning active through time, enabling the continuous adaptation to new conditions.…”

An Adaptive Feedforward Control Method for Under-Actuated Bipedal Walking on the Compliant Ground

“…The second group is based on the relationship between the robot's walking stability and its speed. Specifically speaking, in [22], based on a set of preplanned gaits with different walking speeds, a switched control strategy was developed for the robot prototype ERNIE to realize a stable walking on a treadmill with varying speed; in [23] and [24], by applying the artificial neural network method, the mapping relation between the robot prototype Runbot's walking speed and each joint state was identified, and then the walking stability was improved by regulating the walking speed through the control of each joint motion; in [25], by analysing the effect of robot prototype Meta's three parameters, the amount of ankle push-off, upper body pitch, and step length, on the walking speed, the walking stability was also improved by the control of walking speed. Although the stabilization through the control of robot's walking speed has been realized in some physical experiments, these methods have two drawbacks.…”

Neural control and adaptive neural forward models for insect-like, energy-efficient, and adaptable locomotion of walking machines

“…This is based on a modular structure consisting of different neural modules having main functions that follow three key mechanisms found in animal locomotion (Holst and Mittelstaedt, 1950; Meyrand et al, 1991; Cruse et al, 1998; Katz, 1998; Bläsing and Cruse, 2004; Cruse et al, 2009; Harris-Warrick et al, 2011): (1) central mechanisms [i.e., central pattern generators (CPGs)] for generating basic rhythmic motions, (2) sensory feedback (i.e., afferent-based control) for shaping the motions, and (3) internal forward models (i.e., efferent-based control) for sensory prediction and walking state estimations. While these three key mechanisms are essential for locomotion control as found in biological legged systems, only individual instances of them had been successfully applied to artificial ones (Beer et al, 1997; Ishiguro et al, 2003; Cruse et al, 2007; Kimura et al, 2007; Spenneberg and Kirchner, 2007; Amrollah and Henaff, 2010; Schroeder-Schetelig et al, 2010; Harischandra et al, 2011; Lewinger and Quinn, 2011; Owaki et al, 2012; von Twickel et al, 2012), thereby providing partial solutions. A few studies have applied all these mechanisms to animal-like legged robots to achieve complex behavior and adaptability (Lewis and Bekey, 2002).…”