Modelling capture efficiency of microbots

Abstract

Micro-robotics is one of the fast-growing research domains in the interdisciplinary field of cyber-physical systems. They have immense potential in a plethora of applications like targeted medicine, environmental remediation, micro-scale manufacturing and many more. Major research in this field is being done in the design of micro-robots, predominantly in the propulsion techniques like chemical, optical, magnetic and ultrasonic methods. There is also research going on in parallel in the pure sciences to understand the behaviour of micro-sized particles in the corresponding micro-scale environment, and another parallel track of swarm robots and swarm intelligence to understand the dynamics of natural swarms and to control a group of artificial robots to work collaboratively to achieve a common task respectively.

In the current work, I analyse the capture efficiency of different models of micro-robots into an absorbing target in multiple scenarios of the micro-scale environment. Analytical and simulation modelling of capture efficiency is performed for the simple case of passive non-interacting micro-robots in 2d environment near a square-shaped target region. The variation of the same upon the application of global velocity is also studied. Next, the work looks at capture efficiency of self-propelled micro-robots that can be modelled as active brownian particles (ABPs), run and tumble particles (RTPs) or Chiral ABPs inside circular targets in both 2d and 3d environments in the absence of inter-particle interaction. Further, I focus on the creation of a simulation framework in python that enhances the above-mentioned models of micro-robots with additional features to emulate the complexities of the real-world scenario. These features include inter-particle interaction with each particle's self-propulsion velocity varying with the density of the particles surrounding it; crowded environment where the behaviour of robots near obstacles can either enhance or reduce the capture efficiency; drift velocity that incorporates the inherent drift in the movement of the robots due to the enveloping environment like blood-flow and/or gravity in drug delivery applications or wind-flow in air-borne micro-robots. In the final section, reinforcement learning (RL) technique has been applied for continuous control of artificial micro-swimmers that are magnetically controlled by a Helmholtz coil. Model-free algorithms of RL have been used to learn the navigation of a swarm of magnetic micro-swimmers in different scenarios of reduced state information and varying initial positions. Improved performance has been observed when the state information is modified to include derived observation and also when curriculum learning is used.

This work would be a good adjunct to visualize and verify the theoretical analysis of particles in the micro-scale environment. It can also be a test bed of system-on-chip designs for medical diagnosis and drug tests as it provides a window into the dynamics of micro-particles used there. It is also a simulation environment for data-centric control techniques such as reinforcement learning algorithms to get the swarm of micro-robots (regardless of the propulsion technique) to achieve a particular task under certain constraints, a simple scenario of which is demonstrated in the final chapter of the manuscript. This work serves as the foundation for machine learning engineers to gain domain expertise in the intriguing and highly influential field of micro-robotic systems, positioning them to take a leading role in shaping the future of industrial and commercial applications.