Abstract

Dexterous manipulation is a fundamental yet unresolved topic in robotics, particularly in the realm of micro-manipulation of flexible objects. The complex interplay among external forces, intrinsic material properties, and environmental factors that govern shape deformation presents significant challenges to learning and control. This paper introduces a mechano-flow co-manipulation scheme for flexible micro-objects. It achieves precise handling of floating micro-objects by dynamically adjusting the flow field at the air-liquid interface, enabling simultaneous control of both manipulation and focal planes. A real-time flow prediction framework is proposed based on a reversible neural network, termed FlowNet, achieving bidirectional prediction of input flow and resulting spatio-temporal flow distribution with high accuracy. Through experimental validation, including linear-drive characterization, open-loop trajectory following, and manipulation of different flexible objects, we demonstrate the system’s versatility, precision, and stability. The proposed system offers a promising approach for micro-manipulation through mechano-flow interaction for flexible micro-objects.

Fig. 1 Schematic overview of the mechano-flow co-manipulation system and the target flexible neural electrode. The central illustration depicts the experimental setup, highlighting the water tank geometry, the spatial configuration of the water injection modules, and the resultant surface flow field visualized by streamlines. The magnified insets (a–c) reveal the specific microscopic features of the flexible neural electrode: (a) The proximal end (tail) featuring high-density connection pads. (b) The transition zone displaying the lithographic interconnect traces. (c) The functional distal tip equipped with a specialized “joint hole” designed to facilitate the shuttle-based implantation process.

Fig. 2 Force analysis, motion analysis and experiment screenshots of an electrode floating on the air-liquid interface. (a) Force analysis of a flexible neural electrode floating on the liquid surface under; (i) equilibrium conditions and (ii) unbalanced liquid surface height conditions. (b) In the condition of single-side continuous injection with a flow rate of 1.8 ml/min, the schematic curve of the electrode swing distance in the pixel coordinate system changes with time. The movement is divided into three stages, reverse stage (stage I), forward stage (stage II), and regression stage (stage III). (c) Electrode displacement versus time shows the driving effect of 2 s pulse flow on the electrode. (d) Dynamic equilibrium of the electrode. (e) Experimental screenshots of representative moments in the electrode movement cycle.

Fig. 3 Schematic diagram of network structure and result display. (a) The proposed FlowNet predicts the flow field for the next time step and the input flow based on the flow field over the past T time steps. (b) Angular accuracy vs angular error threshold. (c) Magnitude accuracy vs magnitude error threshold. (d) The real value of the simulation data is compared with the FlowNet network result. The redder and the longer the arrow is, the faster the speed is. (i, ii) vector and scalar ground truth; (iii, iv) vector and scalar results predicted by the first five frames of real value as samples; (v, vi) vector and scalar results predicted by the first five frames as samples and iterated continuously; (vii) scalar prediction results with super-resolution of site size and sampling point density.

https://doi.org/10.1038/s41467-026-77746-z