Research Area "Neurocybernetics"

Research topics

Doctorate Programme
Institute of Cognitive Science
University of Osnabrück
ALEAR (Artificial Language Evolution on Autonomous Robots) 7thFrameworkSmall.png

The EU project ALEAR is an ambitious scientific endeavor that tries to unravel the secrets of cognition and language. Through carefully controlled experiments, in which the complete chain from embodiment to language is modeled, the project partners explore how complex grammatical systems and behaviors can emerge in populations of robotic agents.

Typical Language Games with the Myon Humanoid Robotics Platform (no audio)
typicalLanguageGame1Logo.avi typicalLanguageGame2Logo.avi typicalLanguageGame3Logo.avi
One humanoid chooses a topic to talk about. The second robot performs the action (pointing, moving, putting, ...) it thinks the first robot meant. The first robot confirms (nodding) or declines (head shaking) the action. With many of such experiments and alternating role-play the robots "learn" or "evolve" a collective vocabulary and grammar from scratch. Experiments include action games, spatial games, color games, shape games, and more. Details can be found at the ALEAR Project Website.

All media files of the Neurocybernetics Group for the ALEAR project can be found here.

ALEAR is a joint project of six european institutions:
Full information about the project can be found at the ALEAR Project Website -

The neurocybernetics group works primarily on workpackage WP 3: Behaviour


To support the development of the physical robots morphologies and for the evolution of embodied neural control of humanoid behaviours adequate for the envisaged embodied language games, a physical simulation is necessary. The first objective of this work package is therefore to provide such a simulator for the targeted hardware. It allows the development, study and performance evaluation of the humanoid robots in user-defined environments before everything is fixed in terms of hardware and firmware.

With this simulator basic and advanced behaviours for the humanoids in the project are designed both using evolutionary programming methods and analytic reasoning.

While the hardware is advancing and improved over time, we will further improve the simulator and continue the development of advanced behaviours. This will result in robust and easily-to-reproduce behaviours in any user-defined environment.

Description of Workpackage 3

This work package is divided in three tasks which are closely related to the work performed in work package 2. It will focus on a general simulator for humanoids and its application to the A- and M-series (task 3.1), the co-evolution of morphology and behaviours (task 3.2), and the public release of a library of robust, advanced behaviours for the humanoids (task 3.3).

Task 3.1 Simulator for Humanoids

In this task, we will set up a general physical simulator for humanoids that will allow researchers to easily implement humanoid robots, given its physical data. The simulator will also make it possible to generate various environments with several objects according to the required features that are defined by the user. Among its features, the simulator has an interface with an evolutionary program for evolving the behaviour control of the humanoid and a tool box for on-line analysis of this control. Additionally, there is an interface with the physical humanoid itself that allows back-and-forth switching of the control between simulation and the physical robot. In this way, the researcher can easily compare the simulated behaviour to the actual behaviour of the robot. The end-result of this task will be a hardware infrastructure (workstation and networks) for humanoid simulation and evolution of neural control, and a public general-purpose physical simulator for the research community.

The Simulator for the Humanoids (Top: A-Series, Bottom: Myon)
ORCS_standUpSequence2.AVI ORCS_waistMovement2.AVI ORCS_walk.AVI OASeries_standOnTiltedPlatform.avi
MSeries_Hanging_MotorControl.avi stability.avi Aerobics.avi simpleHandGrasping.avi PDW_Simulation_Cantilever.avi MSeries_MorphologyChanges.avi
More Videos and Descriptions

The subtasks to deliver comprise the following:
  • Measuring of relevant physical characteristics of the robots (first the A-type, later the M-type).
  • Implementing the available proprioceptors and exteroceptors, especially a fast vision simulation.
  • Implementing motor properties including, non-linear models for friction and backlash in kinematical chains of humanoid robots.
  • Implementing body parts and assembling humanoid morphologies.
  • Construction of appropriate physical environments.
  • Adapting an evolutionary program like ESN^3 to the humanoid simulator.
  • Develop and determine problem specific fitness functions and environmental boundary conditions for the embodied language games.
  • Testing evolved controllers on the physical robot.
All subtasks will be performed in close interaction with the embodiment group according to work package 2.

Task 3.2 Co-evolution of Morphology and Behaviour

The co-evolution of morphology and behaviour is one of the most challenging problems in evolutionary robotics. In this task, we will explore bio-inspired mechanisms to generate robust advanced behaviours. This is certainly no trivial task, but the partners involved in this work package have a tremendous amount of know-how and experience in the field with for instance neuroevolution and simulation environment NERD.

Examples of Behaviors for the Language Games (Top: A-Series, Bottom: Myon)
ASeries_walkingFront2_icone.avi ASeries_standShaking.mpg ASeries_gestureLibrary.avi ASeries_pointer.avi ASeries_getUp.avi
GraspingAndPutting_Myon.avi Gestures_Myon.avi Pointing_1D_Myon.avi ArmMovement_MotionCapture2_Myon.avi GraspingObject_MotionCapture3_Myon.avi
PDW_Hanging_Myon.avi CrawlingStandUp_Simulation.avi MSeriesWalking.avi GesturesWithNet.avi Pointing_2D_Myon.avi
More Videos and Descriptions

In the first phase of this task, a library of basic behaviours was made available for the A-series, including standing up, walking, stopping, looking, and pointing. In a second phase, we focus on an optimised morphology for the M-series so that new behaviours become more robust, such as pushing, pulling, grasping, giving, etc. Furthermore, we focus on the following subtasks:
  • Upgrading the evolution-simulator interfaces to allow for co-evolution of morphology and behaviour control.
  • Expanding the physical simulator for new body parts that are needed for advanced language games (e.g. a gripper and additional sensors).
  • Evolving neural control for the advanced behaviours.

Task 3.3 Robustness and Optimisation of Advanced Behaviours

In close cooperation with task 2.3 of work package 2, we will continue the work on robustness and optimisation of advanced behaviours throughout the whole project. The goal of this task is to enhance the behaviours of the robots in the sense that, for example, motion becomes more elegant or walking becomes possible on a rougher terrain. Work will also be performed on the speed of behaviours such as obstacle avoidance and tropisms (going to). Furthermore, depending on the specific properties of the M-series and additional needs of the language game experiments, additional behaviours will be evolved and released in a public library of behaviours.

Involved Staff:
  • Prof. Dr. Frank Pasemann
  • Christian Rempis
  • Verena Thomas
  • Ferry Bachmann
  • Arndt von Twickel