Fig 1 - uploaded by Berend Weel
Content may be subject to copyright.
The Triangle of Life. The pivotal moments that span the triangle and separate the 3 stages are: 1) Conception: A new genome is activated, construction of a new organism starts. 2) Delivery: Construction of the new organism is completed. 3) Fertility: The organism becomes ready to conceive offspring.
Source publication
This paper addresses a principal problem of in vivo evolution of modular multi-cellular robots, where robot 'babies' can be produced with arbitrary shapes and sizes. In such a system we need a generic learning mechanism that enables newborn morphologies to obtain a suitable gait quickly after 'birth'. In this study we investigate and compare the re...
Context in source publication
Context 1
... general, there are two principal forces behind evolution: selection and reproduction. Selection -at least environmental, objective-free selection-is 'for free' in the real world. Therefore, the main challenge for EAE is reproduction, i.e., the creation of tangible physical artifacts with the ability to reproduce. In our case, this means the need for self-reproducing robots. The approach we follow to this end is based on modular robotics with robotic building blocks capable of autonomous locomotion and aggregation into complex 'multicellular' structures in 3D. In this system evolution will not take place in the morphological space of these pre-engineered modules, but in the morphological space of the multicellular organisms. From the perspective of the multicellular robot bodies the basic robots are merely raw material whose physical properties do not change over time. 1 In [6] a conceptual framework for systems where robot morphologies and controllers can evolve in real-time and real-space is presented. This framework, dubbed the Triangle of Life, describes a life cycle that does not run from birth to death, but from conception (being conceived) to conception (conceiving one or more children) and it is repeated over and over again, thus creating consecutive generations of 'robot children'. The Triangle of Life consists of 3 stages, Birth, Infancy, and Mature Life, cf. Fig. 1. In this paper we address a funda- mental problem in the Infancy stage. This stage starts when the morpho- genesis of a new robot organism is completed and the 'baby robot' is delivered. As explained in [6], the body (morphological structure) and the mind (controller) of such a new organism will unlikely fit each other well. Therefore the new organism needs some fine tuning.This problem - the Control Your Own Body (CYOB) problem-is inherent to evolutionary ALife systems where both bodies and minds undergo changes during repro- ...
Citations
... Learners. The RL PoWER implementation follows the description by Jens Kober and Jan Peters [14] and [2] If the organism is from the initial population the algorithm starts by creating the initial policy π 0 with as many splines as there are motors in the organism. These splines are initialised with n values of 0.5 and then adding Gaussian noise. ...
We construct and investigate a strongly embodied evolutionary system, where not only the controllers but also the morphologies undergo evolution in an on-line fashion. In these studies, we have been using various types of robot morphologies and controller architectures in combination with several learning algorithms, e.g. evolutionary algorithms, reinforcement learning, simulated annealing, and HyperNEAT. This hands-on experience provides insights and helps us elaborate on interesting research directions for future development.
... The RL PoWER is reinforcement learning algorithm described by Kober and Peters [?], [6]. The properties of this algorithm were investigated in comparison to HyperNEAT and Simulated Annealing [7], [19]. These studies have shown that RL PoWER is a superior method for on-line gait learning since it converges quickly to learn sufficiently good gaits in a short time. ...
This paper addresses the problem of on-line gait learning in modular robots whose shape is not known in advance. The best algorithm for this problem known to us is a reinforcement learning method, called RL PoWER. In this study we revisit the original RL PoWER algorithm and observe that in essence it is a specific evolutionary algorithm. Based on this insight we propose two modifications of the main search operators and compare the quality of the evolved gaits when either or both of these modified operators are employed. The results show that using 2-parent crossover as well as mutation with self-adaptive step-sizes can significantly improve the performance of the original algorithm.
... Particularly, it requires the generation of rhythmic functions for the activation of the organisms' step motors. The RL PoWER algorithm has been chosen for gait learning in this project based on previous investigations into the infancy phase [29]. ...
... These control points are used for cyclic spline interpolation. The RL PoWER implementation follows the description by Jens Kober and Jan Peters [28] and [29]. If the organism is from the initial population the algorithm starts by creating the initial policy π 0 with as many splines as there are motors in the organism. ...
... Figure 7 shows the learning performance at the start of the infancy and at First we notice that learning does take place: the median performance of organisms at the end of their lifetime is much higher than at the beginning of their lifetime. This is in line with the previous work on our learning algorithm RL PoWER [29]. A new aspect we introduced in this system is the evolution of the mind, to analyse the effect of this evolution we looked at the difference in performance between early and late organisms. ...
This paper presents a software system to conduct investigations into strongly embodied evolutionary robotics, where not only the controllers, but also the morphologies of the robots undergo evolution in real time and real space. The system is based on the recently published 'Triangle of Life' framework and uses the Webots package to simulate physics. The use of a simulator for strongly embodied evolutionary robotics serves two purposes. On the short term it is a proxy until the technology for self-reproducing robots becomes mature. On the long term it is an important part of the future workflow, where it offers a practicable solution for system calibration and quick exploration of the design space without the high costs of using real hardware. Our tool implements an ecosystem of autonomous robots with evolvable bodies and minds that 'live' in an arena simultaneously. In such a world, evolution is not (ab)used to optimise some morphological and/or mental feature in isolation, but employed as a force to adapt all aspects of the robots in a holistic manner. This work forms a stepping stone towards the grand vision of Embodied Artificial Evolution (EAE) or the Evolution of Things as outlined in [1]. The essence of this vision is to construct physical systems that undergo evolution 'in the wild', i.e. not in a virtual world inside a computer. There are various possible approaches towards this goal including chemical and biological ones. The one behind this paper is based on using robots in the broad sense, where the specific substrate is not very important. The bodies can be made of traditional mechatronic components, (self-)assembled from simple modular units, formed by some soft material, 3D printed plastics, some fancy new stuff invented by material scientists, or any combination of these. The important aspect is that we use animate artefacts that can sense, make decisions, and perform actions autonomously. In the system we envision robots are able to actively induce an evolutionary process 'from within' –without a central evolutionary agency– in real time and real space. There are several reasons to be interested in the Evolution of Things. The technology of evolvable robots offers possible applications in the future, where adapting the robot design and/or producing new robots during the operational period without human intervention is important. This can be the case in inaccessible environments, for example, colonies of mining robots that work in extreme depths under the surface of the Earth for extended periods, planetary missions, deep sea explorations, or medical nano-robots acting as 'personal Figure 1: The Triangle of Life. Three pivotal moments span the triangle: 1) Conception: A new genome is activated, construction of a new organism starts. 2) Delivery: Construction of the new organism is completed. 3) Fertility: The organism becomes ready to conceive offspring.
