odNEAT: An Algorithm for Decentralised Online Evolution of Robotic Controllers

Abstract

Online evolution gives robots the capacity to learn new tasks and to adapt to changing environmental conditions during task execution. Previous approaches to online evolution of neural controllers are typically limited to the optimisation of weights in networks with a prespecified, fixed topology. In this article, we propose a novel approach to online learning in groups of autonomous robots called odNEAT. odNEAT is a distributed and decentralised neuroevolution algorithm that evolves both weights and network topology. We demonstrate odNEAT in three multirobot tasks: aggregation, integrated navigation and obstacle avoidance, and phototaxis. Results show that odNEAT approximates the performance of rtNEAT, an efficient centralised method, and outperforms IM-([Formula: see text]), a decentralised neuroevolution algorithm. Compared with rtNEAT and IM-([Formula: see text]), odNEAT’s evolutionary dynamics lead to the synthesis of less complex neural controllers with superior generalisation capabilities. We show that robots executing odNEAT can display a high degree of fault tolerance as they are able to adapt and learn new behaviours in the presence of faults. We conclude with a series of ablation studies to analyse the impact of each algorithmic component on performance.

Automatic Design of Robot Swarms: Achievements and Challenges

“…Experiments were performed in simulation only. Silva et al (2015) introduced odNEAT, a variant of real-time NEAT (Stanley and Miikkulainen, 2002) for the online and decentralized optimization of robot control software. Experiments on three multi-robot tasks (aggregation, navigation with obstacle avoidance, and phototaxis) were performed in simulation only.…”

Decentralized Innovation Marking for Neural Controllers in Embodied Evolution

“…Thus, if a robot moves fast, straight and far from obstacles it will gain energy, and it will lose energy otherwise. This is the same energy update function in which odNEAT was studied [13], and it is inspired from [10].…”

“…However, a global counter as in NEAT's innovation marking mechanism is not directly transferable to Embodied Evolution, because different robots cannot be simultaneously aware of each other's innovations. Two topology-evolving EE algorithms, and more specifically, two mechanisms to mark topological innovations in a distributed manner, have been proposed to date: odNEAT [13] and IM-(µ + 1) [12].…”

“…Recently, odNEAT (Online Distributed NEAT) [13] proposed a distributed version of NEAT in the context of Embodied Evolution. The algorithm uses highresolution clocks at the robot level to mark structural innovations.…”

Embodied Evolution in Collective Robotics: A Review

“…Since the first experiments in this field, the intrinsic online nature of embodied evolution has made such validation comparatively straightforward (Ficici et al, 1999;Watson et al, 2002). "Traditional" evolutionary robotics is more concerned with Silva et al (2012Silva et al ( , 2013Silva et al ( , 2015Silva et al ( , 2017 • with real robots, thus including real-world validation in their research methodology. Since 2010, there have been a number of experiments that employ large (≥100) numbers of (simulated) robots, shifting toward more swarm-like robotics where evolutionary dynamics can be quite different (Huijsman et al, 2011;Bredeche, 2014;Haasdijk et al, 2014b).…”