SwarmPath: Drone Swarm Navigation through Cluttered Environments Leveraging Artificial Potential Field and Impedance Control

TL;DR

SwarmPath tackles the challenge of navigating drone swarms through cluttered environments by coupling a virtual-leader APF global planner with an impedance-based local controller. The method introduces dynamic impedance links that connect follower drones to a virtual leader or to obstacles, enabling smooth deflections and rejoin points to the APF trajectory, thereby maintaining formation connectivity and reducing collision risk. Across simulations and real-world Crazyflie 2.1 experiments, SwarmPath demonstrates improved connectivity and a ~30% reduction in trajectory time, with average trajectory errors around $6$–$7\%$. The work highlights practical potential for rapid, safe swarm navigation in complex settings, with future directions including SLAM-based autonomy and extended applications in forest monitoring and drone racing.

Abstract

In the area of multi-drone systems, navigating through dynamic environments from start to goal while providing collision-free trajectory and efficient path planning is a significant challenge. To solve this problem, we propose a novel SwarmPath technology that involves the integration of Artificial Potential Field (APF) with Impedance Controller. The proposed approach provides a solution based on collision free leader-follower behaviour where drones are able to adapt themselves to the environment. Moreover, the leader is virtual while drones are physical followers leveraging APF path planning approach to find the smallest possible path to the target. Simultaneously, the drones dynamically adjust impedance links, allowing themselves to create virtual links with obstacles to avoid them. As compared to conventional APF, the proposed SwarmPath system not only provides smooth collision-avoidance but also enable agents to efficiently pass through narrow passages by reducing the total travel time by 30% while ensuring safety in terms of drones connectivity. Lastly, the results also illustrate that the discrepancies between simulated and real environment, exhibit an average absolute percentage error (APE) of 6% of drone trajectories. This underscores the reliability of our solution in real-world scenarios.

Figures (9)

• Figure 1: SwarmPath technology performing gate navigation and obstacle avoidance on a complex racing terrain by adaptive impedance link topology where it shows the overall setup of swarm, obstacles and gate position.
• Figure 2: SwarmPath utilizing a swarm of Crazyflie drones, VICON for drone motion tracking, ROS2 for publishing APF path for swarm centroid and impedance parameters for the follower drones. Moreover, PyBullet Environment was used to perform drone simulation.
• Figure 3: Instance of simulation is shown in which drone 3 and drone 4 maintain leader-drone links whereas drone 1 and drone 2 temporarily breaks them in order to avoid collision from gate poles or obstacles. Blue circles of radius $r_{apf}$ represents the APF repulsion field, whereas green circles of radius $r_{imp}$ represents the region where the drone creates a link with an obstacle for deflection. The aim of the drones is to avoid gate poles via adaptive impedance links and reach final position by shortest distance.
• Figure 4: Fully executed SwarmPath trajectory involving four objects; two obstacles and two gate poles (placed in the middle of the obstacles). Drone 3 and drone 4 trails are drawn to show the navigation of agents in challenging scenarios.
• Figure 5: The trajectory generated by conventional APF showing potential connectivity losses among agents.