Commit 27c19873 authored by Matteo De Carlo's avatar Matteo De Carlo

Adding my paper!

parent 3baa3084
Pipeline #131 passed with stage
in 1 minute and 23 seconds
......@@ -3,7 +3,7 @@
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\section{Introduction to Evolutionary Computing}
The work presented in this thesis reflects the result of an ambitious goal to progress the vision of a robotic ecosystem which evolves in real time and real space\cite{2014-invivo, eiben2012embodied, nature-2015}.
The work presented in this thesis reflects the result of an ambitious goal to progress the vision of a robotic ecosystem which evolves in real time and real space\cite{2014-invivo, eiben2012embodied, nature-2015, jelisavcic2017real}.
This implies that the robot morphologies (body, hardware), as well as the controllers (mind, software), are evolvable, i.e., subject to reproduction and selection.
In other words, the primary concern is on robots that can produce offsprings more adaptive to their environment.
......@@ -21,7 +21,7 @@ Last but not least, significant advancements are recently being made in material
All these potential platforms share the ability to create programmable and customised hardware with ease.
Embedding evolution in hardware design is likely going to grow as a much more interesting topic in the future, allowing objects to become more resistant, but also more personalised to the functional and aesthetic needs of the single person.
Over the course of the past six years, a group of researchers at Vrije University Amsterdam have worked on developing the entire system\cite{nature-2015,eiben2012embodied}, in simulation and hardware, as well as its individual components, e.g. particular modules design solutions and specific algorithms for gait learning\cite{Evosphere-2015,jelisavcic2016improving,jelisavcic2017analysis}.
Over the course of the past six years, a group of researchers at Vrije University Amsterdam have worked on developing the entire system\cite{nature-2015,eiben2012embodied}, in simulation and hardware, as well as its individual components, e.g. particular modules design solutions and specific algorithms for gait learning\cite{Evosphere-2015,jelisavcic2016improving,jelisavcic2017real,jelisavcic2017analysis}.
The result is a population of robotic organisms which adapts to the given environment using evolution.
However, the primary concern is still the automation of a robotic learning process, autonomous walking in the first place, specifically gait learning and targeted locomotion.
This thesis represents an attempt to find a possible automated solution for autonomous walking learning that can be implemented in a real-world scenario with physical robots.
......
......@@ -186,7 +186,7 @@ Pseudocode of a possible Spline Controller implementation can be found listed as
% \todo[inline]{list pseudocode here and real code in listing?}
We already used Spline Controllers in the robot baby framework.
The results of those experiments can be found at \cite{jelisavcic2016improving}.
The results of those experiments can be found at \cite{jelisavcic2016improving,jelisavcic2017real}.
In summary, they work quite good within this framework, but they cannot provide any adaptability within the environment because of their open-loop nature.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
......@@ -343,7 +343,7 @@ Their interactions result in iterative improvement of the quality of problem sol
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{\rlpower}
\rlpower \cite{kober2009learning, dangelo2014hyperneat}, abbreviation for \textbf{R}einforcement \textbf{L}ear\-ning with \textbf{Po}licy \textbf{L}ear\-ning by \textbf{W}eighting \textbf{E}xploration with the \textbf{R}eturns, is a reinforced learning algorithm particularly effective for on-line gait development.
\rlpower \cite{kober2009learning, dangelo2014hyperneat, jelisavcic2017real}, abbreviation for \textbf{R}einforcement \textbf{L}ear\-ning with \textbf{Po}licy \textbf{L}ear\-ning by \textbf{W}eighting \textbf{E}xploration with the \textbf{R}eturns, is a reinforced learning algorithm particularly effective for on-line gait development.
The \rlpower implementation follows the description by Jens Kober, Jan Peters \cite{kober2009learning}.
As described by Kober and Peters, the strength of \rlpower is the fact that despite being a reinforcement learning algorithm it does not explore the whole state and action space.
Such extensive exploration would, in fact, require an exorbitant amount of time due to the complexity of a robot.
......@@ -359,8 +359,6 @@ the primary two disadvantages are having a narrow exploration of the search spac
Open-loop controllers cannot react and adapt to the environment, making it easy to be trapped into a local maximum.
Using an open-loop architecture can be useful depending on the task and while being easily trapped in a local maximum seems a problem, sometimes a fast good enough solution is better for the task in hand; especially in such a complex search space like gait development.
%In our previous paper we showed that \rlpower is actually an EA algorithm.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{NEAT}
NeuroEvolution of Augmenting Topologies (NEAT) was first published in 2002 \cite{stanley:ec02} as an innovative method to create ANN exploiting Evolutionary Computing principles without limiting the solution space to a fixed number of nodes or connections.
......
......@@ -23,7 +23,7 @@ The timer is a linear counter that outputs a signal from 0 to 1 in a cyclic mann
A visual representation of the output in a signal over time chart would appear as a sawtooth wave.
The cycle duration is configurable from the experimenter.
In this case, a cycle of 5 seconds was chosen.
Reason being that 5 seconds proved a good value for the \rlpower experiment \cite{jelisavcic2016improving} and therefore it was considered a good starting point.
Reason being that 5 seconds proved a good value for the \rlpower experiments \cite{jelisavcic2016improving,jelisavcic2017real} and therefore it was considered a good starting point.
The \supg neuron has a reset input that, when triggered, makes the neuron prematurely reset his own internal timer cycle, restarting from 0.
The original intent of this input was to be connected with a touch sensor on the feet of the quadruped robot.
......
......@@ -580,6 +580,30 @@ booktitle = "2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016
publisher = "Institute of Electrical and Electronics Engineers, Inc.",
}
@article{jelisavcic2017real,
author = {Jelisavcic, Milan and de Carlo, Matteo and Hupkes, Elte and Eustratiadis, Panagiotis and Orlowski, Jakub and Haasdijk, Evert and Auerbach, Joshua E. and Eiben, A. E.},
title = {Real-World Evolution of Robot Morphologies: A Proof of Concept},
journal = {Artificial Life},
volume = {23},
number = {2},
pages = {206-235},
year = {2017},
doi = {10.1162/ARTL\_a\_00231},
note ={PMID: 28513201},
URL = {
https://doi.org/10.1162/ARTL_a_00231
},
eprint = {
https://doi.org/10.1162/ARTL_a_00231
}
,
abstract = { Evolutionary robotics using real hardware has been almost exclusively restricted to evolving robot controllers, but the technology for evolvable morphologies is advancing quickly. We discuss a proof-of-concept study to demonstrate real robots that can reproduce. Following a general system plan, we implement a robotic habitat that contains all system components in the simplest possible form. We create an initial population of two robots and run a complete life cycle, resulting in a new robot, parented by the first two. Even though the individual steps are simplified to the maximum, the whole system validates the underlying concepts and provides a generic workflow for the creation of more complex incarnations. This hands-on experience provides insights and helps us elaborate on interesting research directions for future development. }
}
@inproceedings{jelisavcic2017analysis,
title={Analysis of Lamarckian evolution in morphologically evolving robots},
author={Jelisavcic, Milan and Kiesel, Rafael and Glette, Kyrre and Haasdijk, Evert and Eiben, AE},
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment