A5005078316- Reinforcement Learning in Robotics
- Modular Robots and Swarm Intelligence
- Evolutionary Algorithms and Applications
- Robot Manipulation and Learning
- Distributed Control Multi-Agent Systems
- Robotic Locomotion and Control
- Evolutionary Game Theory and Cooperation
- Respiratory Support and Mechanisms
- Diffusion and Search Dynamics
- Computability, Logic, AI Algorithms
- COVID-19 Clinical Research Studies
- Gene Regulatory Network Analysis
- Intensive Care Unit Cognitive Disorders
- Sepsis Diagnosis and Treatment
- Non-Invasive Vital Sign Monitoring
- Complex Network Analysis Techniques
- Teaching and Learning Programming
- Neural Networks and Reservoir Computing
- Scientific Computing and Data Management
- Evolution and Genetic Dynamics
- Micro and Nano Robotics
Vrije Universiteit Amsterdam
2020-2024
Delft University of Technology
2020
We generalize the well-studied problem of gait learning in modular robots two dimensions. Firstly, we address locomotion a given target direction that goes beyond typical undirected gait. Secondly, rather than studying one fixed robot morphology consider test suite different robots. This study is based on our interest evolutionary systems where both morphologies and controllers evolve. In such system, newborn have to learn control their own body random combination bodies parents. apply...
The elephant in the room for evolutionary robotics is reality gap. In history of field, several studies investigated this phenomenon on fixed robot morphologies where only controllers evolved. This paper addresses gap a wider context, system both and evolve. context morphology robots becomes variable with currently unknown influence. To examine influence, we construct test suite various evolve their an effective gait. Comparing simulated real-world performance evolved sampled at different...
For mechanically ventilated critically ill COVID-19 patients, prone positioning has quickly become an important treatment strategy, however, is labor intensive and comes with potential adverse effects. Therefore, identifying which intubated patients will benefit may help allocate resources. From the multi-center Dutch Data Warehouse of ICU from 25 hospitals, we selected all 3619 episodes in 1142 invasively patients. We excluded longer than 24 h. Berlin ARDS criteria were not formally...
Designing controllers for robot swarms is challenging, because human developers have typically no good understanding of the link between details a controller that governs individual robots and swarm behavior an indirect result interactions members environment. In this paper we investigate whether evolutionary approach can mitigate problem. We consider very challenging task where with limited sensing communication abilities must follow gradient environmental feature use Differential Evolution...
Abstract Background The identification of risk factors for adverse outcomes and prolonged intensive care unit (ICU) stay in COVID-19 patients is essential prognostication, determining treatment intensity, resource allocation. Previous studies have determined on admission only, included a limited number predictors. Therefore, using data from the highly granular multicenter Dutch Data Warehouse, we developed machine learning models to identify ICU mortality, ventilator-free days ICU-free...
The joint evolution of morphologies and controllers robots leads to a problem: Even if the parents have well-matching bodies brains, stochastic recombination can break this match cause body-brain mismatch in their offspring. This be mitigated by having newborn perform learning process that optimizes inherited brain quickly after birth. An adequate method should work on all possible robot efficient. In paper we apply Bayesian Optimization Differential Evolution as algorithms compare them test...
This study is motivated by evolutionary robot systems where bodies and brains evolve simultaneously. In such `birth' must be followed `infant learning' a learning method that works for various morphologies evolution may produce. Here we address the task of directed locomotion in modular robots with controllers based on Central Pattern Generators. We present bio-inspired adaptive feedback mechanism uses forward model an inverse can learned on-the-fly. compare two versions (a simple...
Natural groups of animals, such as swarms social insects, exhibit astonishing degrees task specialization, useful to address complex tasks and survive. This is supported by phenotypic plasticity: individuals sharing the same genotype that expressed differently for different classes individuals, each specializing in one task. In this work, we evolve a swarm simulated robots with plasticity study emergence specialized collective behavior during an emergent perception Phenotypic realized form...
Legged robots are well-suited for deployment in unstructured environments but require a unique control scheme specific their design. As controllers optimised simulation do not transfer well to the real world (the infamous sim-to-real gap), methods enabling quick learning world, without any assumptions on robot model and its dynamics, necessary. In this paper, we present generic method based Central Pattern Generators, that enables acquisition of basic locomotion skills parallel, through very...
Abstract In this paper, we compare Bayesian Optimization, Differential Evolution, and an Evolution Strategy employed as a gait-learning algorithm in modular robots. The motivational scenario is the joint evolution of morphologies controllers, where “newborn” robots also undergo learning process to optimize their inherited controllers (without changing bodies). This context raises question: How do algorithms when applied various that are not known advance (and thus need be treated without...
Abstract Legged robots, locomoting through ‘limbs’, are well-suited for deployment in unstructured environments. Limbs allow a large range of robot morphologies, with various strengths, but each requiring unique control scheme. As controllers optimized simulation do not transfer well to the real world (the infamous sim-to-real gap), methods enabling quick learning world, without any assumptions on specific model and its dynamics, necessary. In this paper, we present generic method based...
Designing controllers for robot swarms is challenging, because human developers have typically no good understanding of the link between details a controller that governs individual robots and swarm behavior an indirect result interactions members environment. In this paper we investigate whether evolutionary approach can mitigate problem. We consider very challenging task where with limited sensing communication abilities must follow gradient environmental feature use Differential Evolution...
Evolving morphologies and controllers of robots simultaneously leads to a problem: Even if the parents have well-matching bodies brains, stochastic recombination can break this match cause body-brain mismatch in their offspring. We argue that be mitigated by having newborn perform learning process optimizes inherited brain quickly after birth. compare three different algorithms for doing this. To end, we consider algorithmic properties, efficiency, efficacy, sensitivity differences run process.