Target-Search Strategies of Smart Active Agents

Active propulsion allows bacteria and animals to explore their environment and forage nutrients but is also key to the development of future artificial nanoparticles acting as drug delivery agents or able to perform cleansing of soil or polluted water. A central question arising in this context is how smart active agents find their target and how they develop efficient search strategies. In particular, how do agents solve this problem when living in complex environments? Here we will develop reinforcement learning (RL) algorithms to train a smart active particle to find behavioral policies which are optimal to the target-search goal. In RL, agents are trained on a reward and punishment mechanism. The agent is rewarded for correct moves and punished for the wrong ones. In doing so, the agent tries to minimize wrong moves and maximize the right ones. For example, one can use The use of RL to train robots that have the ability to grasp various objects . In our case, we consider smart particles able to change their level of activity (i.e. their self-propulsion speed) and/or other properties of their motion such as for example the persistence of the self-propulsion direction in order to better find their target. The project aims at characterizing and designing optimal strategies adopted by smart active particles to find sparse targets of unknown positions. This problem will be addressed in different environments and we will particularly investigate which are the most effective actions that can be performed by the smart agent. Furthermore, in complex environments, we will identify which characteristics of the environment may provide the most essential cues that the agent can exploit to optimize its target-search strategy. Other questions that we plan to answer concern the transport properties of particles adopting an optimal strategy and the robustness of these strategies with respect to changes in the environment. Our project is mainly focused on the microscopic world, namely, we are interested in natural or artificial swimmers with a typical size of a few micrometers. However, some of the results we will obtain during the realization of this project, are general and then extendible also to larger scales (e.g. animals, drones, or robots).