| 研究実績の概要 |
Our aim was to design controllers for self-organizing swarms of micro-robots with specific collective dynamics emerging from local interactions. In particular, we focused on large robotic swarms with millions micrometer-sized micro-robots composed of sepharose beads controlled through the reaction-diffusion dynamics of DNA strands grafted into the robots. We extended the methodology implemented the previous year to automatically design these DNA-based controllers, by using optimization and quality-diversity algorithms and testing promising solutions in simulations. However this approach was especially computationally expensive, in particular because of the complex reaction-diffusion simulations involved. As such, we presented a new method to improve the computational efficiency of such simulations using GPUs with the CUDA framework (paper "Accelerating the Finite-Element Method for Reaction-Diffusion Simulations on GPUs with CUDA"). Another limitation of our methodology to automatically design robot controllers was that the user previously had to manually provide scores ("feature descriptors") quantifying how diverse the tested solutions were, a typical aspect of quality-diversity algorithms. We described a new methodology to find automatically these diversity scores for a target problem (paper "Ensemble Feature Extraction for Multi-Container Quality-Diversity Algorithms").
|
