MicroPush: A Simulator and Benchmark for Contact-Rich Cell Pushing and Assembly with a Magnetic Rolling Microrobot

Abstract

Magnetic rolling microrobots enable gentle manipulation in confined microfluidic environments, yet autonomy for contact-rich behaviors such as cell pushing and multi-target assembly remains difficult to develop and evaluate reproducibly. We present MicroPush, an open-source simulator and benchmark suite for magnetic rolling microrobots in cluttered 2D scenes. MicroPush combines an overdamped interaction model with contact-aware stick--slip effects, lightweight near-field damping, optional Poiseuille background flow, and a calibrated mapping from actuation frequency to free-space rolling speed. On top of the simulator core, we provide a modular planning--control stack with a two-phase strategy for contact establishment and goal-directed pushing, together with a deterministic benchmark protocol with fixed tasks, staged execution, and unified CSV logging for single-object transport and hexagonal assembly. We report success, time, and tracking metrics, and an actuation-variation measure $E_{Δω}$. Results show that controller stability dominates performance under flow disturbances, while planner choice can influence command smoothness over long-horizon sequences via waypoint progression. MicroPush enables reproducible comparison and ablation of planning, control, and learning methods for microscale contact-rich micromanipulation.

Figures (8)

• Figure 1: System overview of MicroPush. MicroPush consists of an interactive GUI, a physics-based simulator core, interchangeable planning and control modules, a Gym-compatible RL interface, and standardized benchmarking and logging utilities. The modular design supports both interactive prototyping and headless batch evaluation under consistent tasks, metrics, and logging. See Movie S3 for a representative flow-off transport execution.
• Figure 2: Free-space rolling speed versus driving frequency. Markers show the mean speed and error bars indicate $\pm$1 SD across trials. A through-origin linear fit in the low-frequency synchronous range $\omega \le 30$ Hz yields $k_v = 2.3~\mu\mathrm{m/s/Hz}$ used in Eq. \ref{['eq:actuation']}. The vertical dashed line indicates the frequency at which the measured speed reaches its maximum; beyond this point, the speed no longer follows the linear mapping due to step-out effects.
• Figure 3: Geometric quantities for a robot and cell pair used in contact handling and near-field hydrodynamic damping. The center distance is $d_{ij}$, the gap is $h_{ij}=d_{ij}-(r_i+r_j)$, and the overlap is $\delta_{ij}=(r_i+r_j)-d_{ij}$. The unit normal $\mathbf{n}_{ij}$ points from object $i$ to $j$, and the tangential direction $\mathbf{t}_{ij}$ is orthogonal to $\mathbf{n}_{ij}$.
• Figure 4: Global planning in MicroPush. A simulator snapshot is rasterized into a binary occupancy mask. Both planners use the same obstacle inflation model in Eq. \ref{['eq:inflate_radius']}. The weighted A* baseline expands nodes on the inflated grid, whereas AGP operates in continuous geometry and prunes irrelevant obstacles before constructing waypoints.
• Figure 5: Task-level planning for hexagonal assembly. (a) From hexagon center $c$ and radius $\rho$, the planner defines target vertices $\{v_j\}_{j=0}^{5}$ and assigns six cells to vertices (dashed lines). (b) Example execution over the ordered sequence $i_1,\ldots,i_6$: the global planner generates an approach path (blue) to establish contact and a push path (green) to drive the selected cell toward its assigned target vertex; the approach ends at a pre-contact reference point for control.