Review on path planning for obstacle avoidance oriented to micro-/nanorobots

Researchers have conducted a comprehensive study comparing path planning algorithms for navigating micro-/nanorobots through complex and unknown environments. With increasing task complexity, micro-/nanorobots require advanced path planning to ensure safe and efficient movement. Existing algorithms are classified into searching, sampling, and dynamic categories. In future advancements， the deep learning will offer insights into the performance and applicability in varied spatial contexts.

Micro-/nanorobots play a significant role in advancing medical and biological applications due to the robot’s ability to operate at the nanoscale. These tiny robots can perform exact tasks, such as targeted drug delivery and minimally invasive surgery. The small size allows robots to navigate complex environments, reaching areas inaccessible with traditional methods. Micro-/nanorobot’s capability enhances the effectiveness of treatments, reduces side effects, and opens new possibilities for personalized medicine.

Furthermore, micro-/nanorobots are instrumental in research and development, providing insights into cellular and molecular processes, which can lead to breakthroughs in understanding diseases and developing novel therapies. The versatile applications of micro-/nanorobots underscore the importance of revolutionizing healthcare, improving patient outcomes, and advancing scientific research. These applications necessitate precise control over the robots' trajectories and postures, which are achieved via sophisticated path planning algorithms.

Path planning is vital for navigating micro-/nanorobots in complex and dynamic environments. These algorithms help robots determine the optimal path to reach their targets while avoiding obstacles. Choosing the optimal path planning algorithm can enhance task precision, efficiency, and safety when micro-/nanorobots interact with the surrounding environment. The autonomy of micro-/nanorobots permits the undertaking of more intricate tasks without the necessity for human assistance. In general, path planning algorithms are crucial for the successful deployment and functionality of micro-/nanorobots across a wide range of applications.

Therefore, we review the path planning algorithms for micro-/nanorobots, providing a theoretical basis for future research directions and technological breakthroughs. The general process of path planning for micro-/nanorobots includes modeling the environment, generating paths, and modifying paths. According to the environment modeling approach, existing path planning algorithms can be classified into three categories including searching path planning, sampling path planning, and dynamic path planning.

Searching path planning algorithms aim to identify an optimal path by minimizing a cost function while traversing static waypoints, which are modeled from environmental data using grids or pixels. The searching path planning algorithms include the Dijkstra algorithm, the A* algorithm, and the swarm intelligence algorithm. The Dijkstra algorithm, while straightforward in its operational logic, is limited by its high time complexity. In contrast, the A star algorithm provides lower time complexity, but its effectiveness is largely contingent on the design of the cost function. Swarm intelligence algorithms have minimal parameter requirements and high efficiency in conducting global searches.

Sampling path planning algorithms randomly sample waypoints from environmental data. The sampled waypoints provide greater flexibility in the shapes of the micro-/nanorobot’s path, which overcomes the limitations of fixed grids. Sampling path planning algorithms include probabilistic roadmaps and rapidly exploring random trees. The probabilistic roadmap algorithm does not necessitate environmental modeling. However, obtaining the optimal path still requires traversing all sampled waypoints. The rapidly exploring random tree algorithm quickly converges and efficiently discovers paths, though these paths may not be the shortest and often lack smoothness.

Dynamic path planning algorithms simplify global paths into local steps for micro-/nanorobots. Current dynamic path planning algorithms include the dynamic window approach, reinforcement learning, and deep learning. The dynamic window approach algorithm is adaptable but depends on accurate environmental and kinematic modeling. Reinforcement learning algorithms can quickly respond to environmental changes but need distinct training for different scenarios. Deep learning algorithms can autonomously adjust their paths based on environmental inputs but demand extensive datasets for training and operation.

In conclusion, this study provides a comprehensive overview of path planning algorithms, highlighting recent advancements. By understanding the strengths and limitations of each algorithmic approach, researchers can better tailor solutions to complex biomedical challenges, driving innovation in robotic control solutions in the medical field.

Looking forward, the integration of deep learning networks and big data analytics is anticipated to be pivotal in advancing the control and navigation of micro-/nanorobots. These technologies promise to refine path planning further, enhancing the adaptability and efficiency of micro-/nanorobotic systems.

This paper ”Review on path planning for obstacle avoidance oriented to micro-/nanorobots” was published in Robot Learning.