AMARSi (Adaptive Modular Architectures for Rich Motor Skills)



Human and animal movements are still utterly astonishing when compared to robots. The AMARSi Project aims at bridging this gap.

Research includes:

•    analysis and comparison between human motor control and robotics
•    development of damage-robust, safe and fast compliant mechanics
•    exploitation of morphological computation
•    advancing algorithms for unsupervised, reinforcement an imitation learning
•    dynamical and neural models in control architectures across cognitive levels
•    unified framework for locomotion and manipulation behavior

The AMARSi Project has been supported by the FP7 Program, under thematic objective Information and Communication Technology, grant agreement n. 248311

(visit for additional information)



Temporal segmentation of action streams

Movement primitives are usually assumed to extend over short temporal segments only, which begets the question of how these segments can be identified or defined. We developed a motion segmentation algorithm based on Bayesian Binning.

Interaction between periodic and non-periodic kinematic motion primitives
Interaction between periodic and non-periodic kinematic motion primitives

In order to provide a highly controlled setup for the recording of arm movements that are coordinated with walking movements, we have developed a novel virtual reality setup combining motion capture using a VICON system and stereoscopic presentation using a setup with Dolby 3D stereo projectors.  

Identification of motor primitive types

Complex behavior is thought to be generated by a small number of movement primitives. Multiple definitions of motor primitives have been given in the literature, each one translating into different generative models and different techniques for their identifications.


Chiovetto, E., Mukovskiy, A., Reinhart, F., Kansari-Zadeh, M. S., Billiard, A., Steil, J. et al (2014). Assessment of human-likeness and naturalness of interceptive arm reaching movement accomplished by a humanoid robot Perception 43 ECVP Abstract Supplement, page 107. [More] 
Endres, D., Chiovetto, E. & Giese, M. A. (2013). Model selection for the extraction of movement primitives. Frontiers in Computational Neuroscience, 7(185). [More] 
Chiovetto, E. & Giese, M. A. (2013). Kinematics of the coordination of pointing during locomotion. Plos One, 8(11). [More] 
Ajallooeian, M., van den Kieboom, J., Mukovskiy, A., Giese, M. A. & Ijspeert, A. J. (2013). A general family of morphed nonlinear phase oscillators with arbitrary limit cycle shape. Physica D: Nonlinear Phenomena, 263, 41-56. [More] 
Endres, D., Meirovitch, Y., Flash, T. & Giese, M. A. (2013). Segmenting sign language into motor primitives with Bayesian binning. Frontiers in Computational Neuroscience, 7. [More]