Download - Click to edit Master text styles - SDQ Webserver · Click to edit Master title style • Click to edit Master text styles – Second level • Third level – Fourth level G. Hinkel,

Click to edit Master title style

• Click to edit Master text styles – Second level

• Third level – Fourth level

G. Hinkel, H. Groenda, L. Vannucci, O. Denninger, N. Cauli, S. Ulbrich

L’Aquila, 21-JUL-2015

A Domain-Specific Language (DSL) for Integrating Neuronal Networks

in Robot Control




Motivation Platform Python DSL Formal Model Transformation Conclusion

Bio-inspired robots

• Robots inspired by nature

• Desire to create biologically plausible robot controllers

21.07.2015 Georg Hinkel et al. - A Domain-specific Language for Integrating Neuronal Networks in Robot Control 2

[Roe14]





Connecting Robots to Neuronal Networks


?

[Image: extremetech.com]

[Image: FZI]





The Human Brain Project

• Unify understanding of the (human) brain – Provide ICT platforms

• Transfer this knowledge into products – Future Neuroscience

– Future Medicine

– Future Computing

• Interdisciplinary cooperation






The Neuro-Robotics Platform (NRP)

Closed Loop

Engine

Transfer Functions

World Simulator (Gazebo)

Transfer Functions

Neuronal Simulator

(Nest)


Camera images, Sensors

Control messages Membrane potentials

Network stimuli





The NRP in action






A basic example experiment


[Brai86] [Image: Clearpath robotics]

EyeSensorTransmit

WheelSensorTransmit





Components of a running simulation






Designing a DSL

• General-purpose language inappropriate [Fow10]

• Developers do not voluntarily change their primary language [MR13]

• Python popular among neuroscientists & roboticists – Goal to make DSL familiar to Python developers

• No limit for expressiveness in the Transfer Functions – Internal Python DSL






A Transfer Function for the example

from geometry_msgs.msg import Vector3, Twist

def wheel_transmit(t, left_wheel_neuron, right_wheel_neuron):

linear = Vector3(20 * min(left_wheel_neuron.voltage, right_wheel_neuron.voltage), 0, 0)

angular = Vector3(0, 0, 100 * (right_wheel_neuron.voltage - left_wheel_neuron.voltage))

return Twist(linear=linear, angular=angular)


Import Transfer Functions Framework

Mark Python Function as Transfer Function

Mapping of parameters to neurons

import hbp_nrp_cle.tf_framework as nrp

@nrp.MapSpikeSink("left_wheel_neuron",

nrp.brain.actors[0], nrp.leaky_integrator_alpha)

@nrp.MapSpikeSink("right_wheel_neuron",


@nrp.Neuron2Robot(Topic('/husky/cmd_vel', Twist))





Approaching a formal model

• Neuroscientists often do not have programming expertise

• Efficient failure management – Decreased time for errors reporting

– Increased locality for errors

• Abstract syntax for graphical editor






Combination of DSLs on multiple abstraction levels






Using the formal model

• Transformation of Transfer Functions – Simulation component (CLE) only needs single

input format

– Expert users can trace the transformation

• Formal model may contain references to Transfer Functions in Python DSL






A Transfer Function for the example


import hbp_nrp_cle.tf_framework as nrp

from geometry_msgs.msg import Vector3, Twist

@nrp.MapSpikeSink("left_wheel_neuron",


@nrp.MapSpikeSink("right_wheel_neuron",


@nrp.Neuron2Robot(Topic('/husky/cmd_vel', Twist))

def wheel_transmit(t, left_wheel_neuron, right_wheel_neuron):

linear = Vector3(20 * min(left_wheel_neuron.voltage, right_wheel_neuron.voltage), 0, 0)

angular = Vector3(0, 0, 100 * (right_wheel_neuron.voltage - left_wheel_neuron.voltage))

return Twist(linear=linear, angular=angular)





Conclusion

• Internal DSL in Python for Transfer Functions – Raises abstraction level of the platform

– Familiar for Python users

• Formal model for integrating neuronal networks with robot control – Advantages in model validation

– Enabling artifact for graphical editor

• Support of both abstraction levels – Transformation approach

– Allow expert users to use Python DSL






Outlook

• Validation in more realistic scenarios – Platform will go live soon

– Collection of user feedback

• Graphical DSL – Abstract syntax already existing

– Support users with little programming expertise





Thank you for your attention


[email protected]





Conclusion

• Internal DSL in Python for Transfer Functions – Raises abstraction level of the platform

– Familiar for Python users

• Formal model for Integrating neuronal networks with robot control – Advantages in model validation

– Enabling artifact for graphical editor

• Support of both abstraction levels – Transformation approach

– Allow expert users to use Python DSL





References

[Roe14] A. Roennau, G. Heppner, M. Nowicki, and R. Dillmann, „LAURON V: a versatile six-legged walking robot with advanced maneuverability,” in Advanced Intelligent Mechatronics (AIM), 2014 IEEE/ASME International Conference on, IEEE, 2014, pp. 82-87.

[Völ06] M. Völter, T. Stahl, J. Bettin,A. Haase, S. Helsen, and K. Czarnecki, Model-Driven Software Development: Technology. 2006

[Brai86] V. Braitenberg, Vehicles: Experiments in synthetic psychology. MIT press, 1986

[MR13] L. A. Meyerovich and A. S. Rabkin, “Empirical analysis of programming language adoption” in Proceedings of the 2013 ACM SIGPLAN International Conference on Object Oriented Programming Systems Languages & Applications, ACM, 2013, pp. 1-18

[Fow10] M. Fowler, Domain-specic languages. Pearson Education, 2010.

Download - Click to edit Master text styles - SDQ Webserver · Click to edit Master title style • Click to edit Master text styles – Second level • Third level – Fourth level G. Hinkel,

Top Related