Systems, frameworks, and methods for reinforcement learning (RL)-based real-time path planning for unmanned aerial vehicles (UAVs) are provided. The learning capabilities of federated learning (FL) can be integrated with an improved deep RL framework, including using a significant reply memory buffer (SRMB) to accelerate the intelligent behavior. The framework can train a UAV to intelligently dodge static and dynamic defense systems and achieve assigned goals (e.g., in a hostile area). The FL can enable collaborative learning through a swarm of UAV agents.

A system for guiding a drone to an intended destination using a remote guidance system, independent of a global positioning system installed on the drone and independent of radio guidance. The system uses a two-way optical communication channel between the guidance system and the drone. The drone and the guidance system each have a light source emitting a beam of encoded light, such as a modulated laser beam, and having an extended field of illumination, and a detector receiving the impinging light beam. The guidance system can detect the angular location of the drone emission, and can transmit instructions optically to the drone, while the drone can receive flight path instructions from the guidance system. The drone can be launched from a position that is not in the line of sight of its intended destination and guided optically from the launch position to its intended target destination.