Videos

EPFL iCubDemo2

cCub Dancing

EPFL iCubDemo

cCub First Walking Steps

Human crawling

Motor primitives provide a novel perspective on the neural control system for locomotion by revealing fundamental motor activation patterns that can account for the varied and complex activations of specific muscles. While the modular organization of motor patterns has been received considerable attention during last years, still, the development of motor primitives represents an intriguing topic and has not been studied previously. Motor primitives may reflect in some way how the nervous system develops, by building up or modifying modules as it matures. One way to examine this possibility is to examine the motor primitives underlying infant stepping, toddler and adult walking. Babies can produce coordinated stepping shortly after birth, but the behavior is lost within a few weeks, only to reappear 6-8 months later when it evolves first into crawling and then into walking. By studying infant crawling and its comparison with motor patterns of adults and quadrupedal animals we aim at uncovering a common underlying neural framework for the modular control of human locomotion and its development.

 

FlexIRob (Flexibel Interactive Robot)

A showcase scenario developed at Bielefeld University to combine a compliant robot (Kuka-LWR), kinesthetic teaching in user interaction for data recording, and motion learning with a reservoir computations approach in a coherent robot control architecture. The compliant robot interactively learns to use different redundancy resolution in different parts of the workspace, whereas the postures to be used are defined by the data recorded in the respective workspace parts respectively. The reservoir network then extrapolates the solution to untrained areas of the workspace. The system is also a demonstration of the tight interconnection of interaction, motion learning, and compliance in a coherent software architecture that is based on a memory based middleware earlier developed at Bielefeld University.

Cheetah quadruped compliant robot

The simulated Cheetah quadruped robot (simulation environment: Webots) performs a trot gait, with a joint control signal oscillating at frequencies up to 3.5Hz. Signal parameters are found through systematic testing. The cost function includes a mixture of speed and stability, for the latter we are measuring rolling and pitching of the robot body. The Cheetah simulation model is developed at Biorob/EPFL in the context of the AMARSI EU project. It is distributed among partners of the AMARSI project to provide a simulator environment to test and analyse different robot locomotion control approaches (e.g. central pattern generators, dynamical movement primitives, reservoir computing) to reach a higher level of robotic motor skills. The Cheetah robot was developed at Biorob/EPFL (2008), it is a small bio-inspired, compliant quadruped robot with three-segment legs. An enhanced version will be designed and implemented as a common platform for the AMARSI partners.

Catching

Intercepting a moving object is a demanding motor skill because visual stimuli guiding movement execution change in time and the central nervous system (CNS) must update on-line or predict the object position to compensate for sensorimotor latencies and bring the arm to the right place at the right time. How the human control architecture combines visual information and prior knowledge of physical laws to guide interceptive actions coordinating many degrees-of-freedom is still an open question. To date, several experimental paradigms have been used to investigate the sensorimotor control mechanisms employed by the CNS to accomplish a variety of interceptive tasks but very few of them have investigated the interception of balls flying in three dimensional space because it is generally difficult to control the ball flight parameters when air drag cannot be neglected. To systematically investigate naturalistic interceptive tasks we have developed and used a novel actuated launching system that allows to control the flight time and arrival position of a ball projected from a fixed location and to track the ball trajectory throughout its flight. Surprisingly, in contrast to stereotypical characteristics of reaching movements towards stationary objects, motion capture data reveal large inter-individual differences in arm kinematics for catching flying balls with the same flight parameters.