Leveraging Online Racing and Population Cloning in Evolutionary Multirobot Systems

Silva, Fernando; Correia, Luís; Christensen, Anders Lyhne

doi:10.1007/978-3-319-31153-1_12

Cited by 7 publications

(12 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While the online evolutionary process is able to consistently synthesize a set of solutions to the homing task, there are different options to potentially increase the performance of evolution in terms of the percentage of successful controllers. In recent simulation-based contributions, we have developed two complementary approaches, namely: (i) a racing technique to cut short the evaluation of poor controllers based on the task performance of past controllers [ 29 ] and (ii) a novel paradigm called online hyper-evolution [ 30 , 31 ], which can both combine the benefits of different algorithms for controller generation over time, and construct algorithms by selecting which algorithmic components should be employed for controller generation (e.g. mutation, crossover, among others).…”

Section: Discussionmentioning

confidence: 99%

Evolutionary online behaviour learning and adaptation in real robots

2017

Self Cite

View full text Add to dashboard Cite

Online evolution of behavioural control on real robots is an open-ended approach to autonomous learning and adaptation: robots have the potential to automatically learn new tasks and to adapt to changes in environmental conditions, or to failures in sensors and/or actuators. However, studies have so far almost exclusively been carried out in simulation because evolution in real hardware has required several days or weeks to produce capable robots. In this article, we successfully evolve neural network-based controllers in real robotic hardware to solve two single-robot tasks and one collective robotics task. Controllers are evolved either from random solutions or from solutions pre-evolved in simulation. In all cases, capable solutions are found in a timely manner (1 h or less). Results show that more accurate simulations may lead to higher-performing controllers, and that completing the optimization process in real robots is meaningful, even if solutions found in simulation differ from solutions in reality. We furthermore demonstrate for the first time the adaptive capabilities of online evolution in real robotic hardware, including robots able to overcome faults injected in the motors of multiple units simultaneously, and to modify their behaviour in response to changes in the task requirements. We conclude by assessing the contribution of each algorithmic component on the performance of the underlying evolutionary algorithm.

show abstract

Section: Discussionmentioning

confidence: 99%

Evolutionary online behaviour learning and adaptation in real robots

2017

Self Cite

View full text Add to dashboard Cite

show abstract

“…Additionally, the robots used a range and bearing system to detect the presence of neighbouring robots. Although Silva et al (Silva et al, 2015;Silva et al, 2016) also evolve artificial neural network controllers in order to carry out phototaxis, they do so through online learning. These experiments use a distributed and decentralised neuroevolution algorithm that continuously modifies robots' behaviours in order to respond to changes and unforeseen circumstances.…”

Section: Related Workmentioning

confidence: 99%

Evolving collective behaviours in simulated kilobots

Holland

Griffith

O’Riordan

2018

Proceedings of the 33rd Annual ACM Symposium on Applied Computing

View full text Add to dashboard Cite

“…This approach was successful in finding solutions in one hundred or less generations, but it failed to scale successfully if hundreds of generations were required. Silva et al [14] carried out a similar ap-proach, whereby they cut short the evaluation of poor controllers which proved effective in experiments with a high selective pressure, but arguably decreased diversity amongst the population.…”

Section: Related Workmentioning

confidence: 99%

Identifying the Reality Gap Between Abstract and Realistic Models Using Evolved Agents and Simulated Kilobots

Holland¹,

Gallagher²,

Griffith³

et al. 2018

2018 IEEE International Conference on Robotics and Biomimetics (ROBIO)

View full text Add to dashboard Cite

A common challenge in evolutionary swarm robotics is the transfer of simulated results into real-world applications. This difficulty can arise in a variety of real-world settings and problems such as sensory differences in robots and changes in the environment. We identify this reality gap at a simulation level by comparing the evolved behaviours of simulated Kilobots in two different models with different levels of abstraction. Our aim is to identify the reality gap that occurs at simulation level by increasing the task difficulty and noting differences in outcomes. Insights gained in this process may help rule out any further causes of reality gap when moving to experiments with physical robots.

show abstract

Leveraging Online Racing and Population Cloning in Evolutionary Multirobot Systems

Cited by 7 publications

References 19 publications

Evolutionary online behaviour learning and adaptation in real robots

Evolutionary online behaviour learning and adaptation in real robots

Evolving collective behaviours in simulated kilobots

Identifying the Reality Gap Between Abstract and Realistic Models Using Evolved Agents and Simulated Kilobots

Contact Info

Product

Resources

About