The rise of e-commerce has driven a significant increase in market demand for robust and cost-effective logistics solutions. This demand is high in both central and rural areas, where last-mile delivery faces additional challenges to reach end customers. As a result, many companies are exploring new ways to streamline their supply chains and reduce delivery times. Autonomous vehicles (aerial and ground), with their ability to reach remote locations and lower human operational costs, have become a potential solution. In this direction, several key logistic and aerospace companies have come forward to develop aerial and ground vehicles for logistic deliveries. 

The main objective of this project is to develop an advanced system driven by cutting-edge Deep Learning (DL) AI algorithms to achieve a high level of autonomy and versatility with a swarm of aerial and ground vehicles in logistics. We intend to compare different approaches to task dispatching based on multi-agent DRL, as well as on the novel GALOPP architecture and natural swarm intelligence. These approaches will enable cooperative route planning for multiple drone and ground robot deliveries, ensuring efficient and coordinated operations within the swarm. To navigate successfully and avoid collisions the swarm will generate a route for each robot and avoid dynamic collisions while passing this route. The multi-agent localization and path planning approach will be developed for a dynamic swarm behavior. To achieve a precise swarm docking on a ground vehicle in the presence of visual occlusion, wind force and ground effect, we propose a novel approach based on DRL. Our landing algorithm will include platform detection, motion prediction, and landing behavior for both static and mobile platforms. We will employ DRL agents, starting with the A2C or PPO model. The proposed fundamental approach will rely on the heterogeneous capability of the swarm, where drones with different roles and tools will imply different policies to simultaneously achieve the delivery in an unstructured environment. This scenario should increase the scalability and sustainability of the system compared with the existing solutions. 

Several multi-agent delivery experiments will be conducted in the indoor and outdoor environments with scaling the swarm size. 

"RSF Grant 24-41-02039. LogiSAR: A Heterogeneous Swarm of Autonomous Robots with Deep Learning for Developing the Next Generation of Smart Logistics. RSF-DST 2024 Competition 'Conducting Fundamental Scientific Research and Exploratory Scientific Research by International Scientific Teams'."