Microrobot Vascular Parkour: Analytic Geometry–based Path Planning with Real-time Dynamic Obstacle Avoidance

Yanda Yang, Max Sokolich, Fatma Ceren Kirmizitas, Sambeeta Das, Andreas A. Malikopoulos · 2025

Read Microrobot Vascular Parkour: Analytic Geometry–based Path Planning with Real-time Dynamic Obstacle Avoidance by Yanda Yang, Max Sokolich, Fatma Ceren Kirmizitas, Sambeeta Das, Andreas A. Malikopoulos (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-05500.txt.

Introduction

Tiny magnetic robots that steer themselves through living tissue are demonstrated by Yanda Yang and five co-authors in a 2025 robotics paper. Such microrobots could enable gentler interventions in tight biological spaces, but they must navigate moving obstacles and flowing fluids in real time.

Their system splits the job: a geometry-based planner draws safe long-range routes from microscope images, while local controllers, rule-based or learned, dodge nearby movers before handing control back. The planner produces short paths far faster than standard alternatives, inside the video-rate budget. Tests cover microfluidic channels with moving particles and blood cells, flow with and against the robot, plus actuation in eye and brain tissue samples.

Catalog record

CREATORYanda Yang, Max Sokolich, Fatma Ceren Kirmizitas, Sambeeta Das, Andreas A. Malikopoulos
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-05500

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Microrobot Vascular Parkour: Analytic Geometry–based Path Planning with Real-time Dynamic Obstacle Avoidance Yanda Yang, Max Sokolich, Fatma Ceren Kirmizitas, Sambeeta Das, Andreas A. Malikopoulos Abstract Autonomous microrobots in blood vessels could enable minimally invasive therapies, but navigation is challenged by dense, moving obstacles. We propose a real-time path planning framework that couples an analytic geometry global planner (AGP) with two reactive local escape controllers, one based on rules and one based on reinforcement learning, to handle sudden moving obstacles. Using real-time imaging, the system estimates the positions of the microrobot, obstacles, and targets and computes collision-free motions. In simulation, AGP yields shorter paths and faster planning than weighted A* (WA*), particle swarm optimization (PSO), and rapidly exploring random trees (RRT), while maintaining feasibility and determinism. We extend AGP from 2D to 3D without loss of speed. In both simulations and experiments, the combined global planner and local controllers reliably avoid moving obstacles and reach targets. The average planning time is 40 ms per frame, compatible with 25 fps image acquisition and real-time closed-loop control. These results advance autonomous microrobot navigation and targeted drug delivery in vascular environments. Figure 1 : Framework for microrobot navigation and dynamic obstacle avoidance in blood vessels. (A) Real-time video processing and extracting microrobot and obstacle states, including safety zones. (B) Analytic geometry planner (AGP) computes a collision-free path to the target. When the microrobot enters a safety zone, control switches to the local escape module. (C) Extension to three dimensions: mapping from the three-dimensional environment to the two-dimensional imaging plane for planning and lifting the planned path back to three dimensions. (D) Local escape controllers: a rule-based controller and a reinforcement learning-based controller. When the microrobot escapes a safety zone, AGP replans the global path. (E) The microrobot moves downstream and upstream in a blood vessel while avoiding moving red blood cells. INTRODUCTION Microscale robotics is emerging as a platform for precise, minimally invasive therapies, especially targeted drug delivery [ 1 , 2 ] . Conventional delivery routes (oral or intravenous) often lack spatial specificity, leading to off-target effects and reduced efficacy [ 3 ] . Recent magnetic and bio-inspired microrobots can carry therapeutic payloads directly to sites of disease (e.g., tumors, inflamed tissue, occlusions) [ 4 , 5 , 6 ] . Their small size enables navigation in confined spaces with noninvasive actuation and precise targeting, simplifying otherwise complex procedures [ 7 , 8 ] and opening avenues in regenerative medicine [ 9 ] . Because micrometer agents cannot house actuators, sensors, or processors onboard, they are propelled and steered by external fields. A broad set of actuation modalities has been explored: magnetic and electric fields, chemical reactions, acoustics, and light-responsive materials [ 10 , 11 , 12 , 13 , 14 , 15 ] . Chemical actuation can raise biocompatibility concerns, and optical methods are limited by penetration depth in tissue. Magnetic actuation avoids these constraints and has therefore become a leading choice for biomedical microrobotics [ 16 , 17 , 18 , 19 , 20 ] . Effective use of magnetic microrobots depends on reliable path planning and obstacle avoidance in blood vessels [ 21 , 22 , 23 , 24 , 25 , 26 ] . Existing planners span optimization (e.g., particle swarm optimization, PSO) [ 27 , 28 ] , graph search (e.g., A ∗ ) [ 29 , 30 ] , and sampling (e.g., rapidly exploring random trees, RRT) [ 31 , 32 , 33 , 34 ] . Each has drawbacks for time-critical navigation: PSO can converge slowly and inconsistently due to multi-particle exploration; A ∗ can incur heavy search costs in large or high-dimensional grids; and RRT is stochastic, with variable quality and no optimality guarantee. A central challenge is real-time planning in the presence of dynamic obstacles [ 35 , 36 ] . In blood vessels, microrobots frequently encounter moving red blood cells, protein aggregates, and other debris, demanding rapid, reactive decisions. Some dynamic obstacle avoidance methods have been developed in recent years, such as radar-based methods [ 37 , 38 ] , artificial potential fields [ 22 ] , Dynamic Window Approach (DWA) [ 39 ] , fuzzy logic [ 40 ] , and learning-based policies [ 41 , 42 ] . But there are still some unresolved issues. Extensions of traditional path planning methods often fail to meet real‐time computational requirements due to inherent algorithmic complexity. Currently, most prior work typically targets low-speed swarms, and real-time planning for high-speed rolling microrobots remains underexplored. Moreover, most prior work evaluates their methods in simulated or single‐scenario environments with a simple dynamic obstacles setting, which limits insight into how these methods perform in realistic, complex, dynamic settings and prevents evaluation of their generalization across diverse conditions. To address these issues, we introduce a real-time framework for autonomous microrobot navigation (Figure 1 ). A deterministic analytic geometry–based global planner (AGP) produces a collision-free path without search or sampling. Two complementary local escape controllers, one rule-based and one reinforcement learning–based, handle sudden, short-range interactions with moving obstacles. The system processes a live microscope video stream into frames, generates masks from them, and identifies contours to extract details such as obstacle dimensions, safe zones, and location coordinates. Given the microrobot’s current position and a predefined target point, the AGP calculates the near-optimal collision-free path based on the latest frame information. When a dynamic obstacle appears and the microrobot enters the safety zone of the obstacle, the local escape controller activates and takes over. Once the microrobot successfully escapes the safe zone, the AGP replans the global path. This iterative process continues until the target is reached. The contributions of our work are as follows: 1. A real-time navigation framework that couples a deterministic global planner (AGP) with two reactive local controllers (rule-based and reinforcement learning-based). We validate the system in simulation and vascular-like microfluidic environments with realistic dynamic obstacles. 2. A fast, deterministic global planner: AGP yields feasible paths and outperforms WA ∗ , PSO, and RRT in path length and planning time. With the local controllers, the average planning time is 40 ms per frame, satisfying closed-loop real-time operation. 3. A 3D extension of AGP that preserves computational speed, enabling real-time planning in complex spatial geometries. 4. A reinforcement learning controller dedicated to short-range dynamic avoidance. Restricting learning to this local task improves sample efficiency and robustness, while the modular design keeps global plans interpretable and easily re-plannable. 5. A deterministic intermediate waypoint controller based on relative microrobot and obstacle geometry. It requires no training, is human-readable and verifiable, and provides predictable responses in scenarios where learned policies might generalize poorly. We first evaluate the framework across diverse dynamic simulations, including vascular phantoms, to assess versatility and robustness. We then integrate it with a magnetic control system and demonstrate collision-free navigation in microfluidic chips containing dynamic obstacles (silica particles and human red blood cells), validating reliable operation under realistic conditions. RESULTS Environmental Information The cell–microrobot environment can be known by fluorescence or bright-field imaging. From each frame of the real-time video stream, image processing extracts the positions and sizes of all objects in the environment. For every frame, we convert the image from BGR to HSV and to grayscale

Read the complete text