Autonomous environment-adaptive microrobot swarm navigation enabled by deep learning-based real-time distribution planning

Navigating a large swarm of micro-/nanorobots is critical for potential targeted delivery/therapy applications owing to the limited volume/function of a single microrobot, and microrobot swarms with distribution reconfigurability can adapt to environments during navigation. However, current microrobot swarms lack the intelligent behaviour to autonomously adjust their distribution and motion according to environmental change. Such autonomous navigation is challenging, and requires real-time appropriate decision-making capability of the swarm for unknown and unstructured environments. Here, to tackle this issue, we propose a framework that defines different autonomy levels for environment-adaptive microrobot swarm navigation and designs corresponding system components for each level. To realize high autonomy levels, real-time autonomous distribution planning is a key capability for the swarm, regarding which we show that deep learning is an enabling approach that allows the microrobot swarm to learn optimal distributions in extensive unstructured environmental morphologies. For real-world demonstration, we study the reconfigurable magnetic nanoparticle swarm and experimentally demonstrate autonomous swarm navigation for targeted delivery and cargo transport in environments with channels or obstacles. This work could introduce computational intelligence to micro-/nanorobot swarms, enabling them to autonomously make appropriate decisions during navigation in unstructured environments.