Edge Artificial Intelligence

24CDLPP: Optimized Multi-Drone Path Planning in Evolving Environments Using Machine Learning for Object Detection

Lade Gunakar Rao* and K. Rajchandar

School of CSE (CS & AI), SR University, Warangal, Telangana, India

Abstract

One of the significant challenges in multi-drone path planning is obstacle avoidance combined with real-time navigation. Most conventional path-planning algorithms face difficulties adapting to environments that evolve in real-time, where an obstacle may emerge or be moved at any given time, without knowledge beforehand of its location. We thus suggest an optimized multi-drone path-planning framework based on machine learning techniques that involve real-time object detection and navigation in the evolving environment. Integration of deep learning-based object detection systems, including CNNs, into drones can be made to enable the detection and classification of obstacles with high accuracy. This updates the navigation algorithm continuously so that the drone dynamically changes its path. The hybrid approach utilizes machine learning models and traditional optimization algorithms. This allows for realizing exploration-exploitation trade-offs under time constraints. The approach will enable UAVs to decide on their own during such real-time situations, and hence nurture the ability of drones to research complex ground scrapes affected by different degrees of obstacles and variations in the environment. We validate the proposed method via simulations and ...