A Microscopic Vision-Based Robotic System For Floating Electrode Assembly

The implantation of multichannel, miniaturized, flexible neuroelectrodes for high-quality brain signal acquisition is of great importance for brain science research and brain–computer interfacing (BCI). However, slender and thin flexible neuroelectrodes usually require a tungsten probe as the shuttle to assist in penetrating the pia mater for implantation. The process in which the tungsten probe passes through the engaging hole on the tip of the electrode and is tightly bonded is called electrode assembly, which is challenging due to the small-scale and fragile microstructures. The conventional manual assembly is error-prone and time-consuming with low yields. It has a high risk of electrode damage, requiring extensive training, very stable hand– eye coordination, and a high level of manual dexterity of the operator. The development of a robot-controlled microassembly system is essential for neuroscience research and clinical deployment. This article presents a universal automated microscopic vision-guided robotic system for brain electrode assembly. A robot system with learning-based detection combined with visual servoing is developed for 3-D object and pose estimation, and a robot with submicron displacement accuracy achieves the precise control of the probe. In addition, a new end-to-end deep learning network is designed for microfeature detection, and a palpation-based motion strategy is proposed to enable motion control with missing depth information in the microenvironment. Detailed experiments are performed on actual BCI electrodes, and the entire process is automated with high efficiency with an average assembly time of 32.2
$\pm$
6.4 s and a successful rate of over 90%.