The ParcelDrone project involves the design and implementation of a modular aerial delivery system utilizing a Graph Neural Network (GNN)-based decentralized control strategy. The system is engineered to operate autonomously, accommodating parcels of varying shapes, dimensions, and masses while maintaining scalability and control redundancy. The modular design of ParcelDrone ensures flexibility, enabling collective 6 Degree-of-Freedom (6 DoF) motion through thrust differentials generated by an assembly of independent modules. Unlike centralized controllers, which are limited in scalability and prone to reduced performance under fault conditions, ParcelDrone employs a decentralized approach. Each propeller module functions as an independent agent within a multi-agent framework, ensuring robust control redundancy. To achieve seamless scalability without requiring modifications to the control architecture, an aggregation Graph Neural Network (aggregation GNN) is used. This innovative network trains each module to calculate its control effort based on its local states and the parcel’s state information, eliminating the need for inter-module communication. Extensive experimental trials have validated the system's capabilities, demonstrating its effectiveness in hover stability and trajectory tracking tasks. This project represents a significant advancement in modular aerial robotics, offering a scalable, fault-tolerant solution for parcel delivery.

This project focuses on designing a visual servoing strategy that integrates a Physics-Informed Neural Network (PINN) with a dynamics-centered visual servoing technique for multi-rotors. The approach aims to address system uncertainties, inaccuracies, and modeling challenges without the need for inverse Jacobian calculations. By directly relating pixel variations to the torque and thrust inputs of the multi-rotor, this method simplifies the control process while enhancing robustness. The PINN is employed to model and mitigate uncertainties in camera and multi-rotor parameters as well as to address the inaccuracies inherent in dynamics-centered visual servoing techniques. Compared to state-of-the-art data-driven methods, the PINN-based strategy requires 65% less labeled data for characterizing uncertainties and inaccuracies, making it highly efficient for practical deployment. To ensure real-time performance, the PINN-learned model is combined with an adaptive horizon monotonically weighted nonlinear model predictive controller (NMPC). This controller achieves control effort processing rates 10 times faster than conventional Tube MPC and Adaptive MPC approaches, enabling efficient trajectory tracking in real-world scenarios. The effectiveness of this strategy has been demonstrated through real-time trajectory tracking experiments, showcasing its ability to approximate modeling inaccuracies and handle uncertainties of up to 70% in camera parameters. This project represents a significant step forward in robust and efficient visual servoing for multi-rotor systems.

This project focuses on the creation of a depth-independent augmented dynamics visual servoing method for multirotors, overcoming key limitations of monocular camera-based visual servoing by eliminating the dependency on depth information. The proposed approach establishes a direct relationship between pixel variations and the multirotor’s thrust and torque commands, enabling precise control without the need for inverse Jacobian computations. The visual servoing strategy enhances robustness by allowing the outer loop controller to effectively manage image noise within system uncertainty. This reduces the need for extensive image filtration, thereby improving the system’s resilience to noise. To regulate the multirotor’s augmented dynamics, a super-twisting fast terminal sliding mode controller is employed, ensuring finite-time error convergence with minimal chattering in control efforts. The performance of this method has been validated through numerical simulations and real-time experiments, demonstrating its effectiveness in trajectory tracking and control. Furthermore, the technique has been benchmarked against traditional position and image-based visual servoing approaches, showcasing its superior accuracy and robustness. This project offers a significant advancement in the field of visual servoing for multirotors, particularly in scenarios where depth information is unavailable or unreliable.

This project addresses the challenge of maintaining control and stability in hexacopters with alternating rotor configurations during rotor failures. Rotor failure introduces asymmetric propulsion, causing residual moments along the arm of the failed rotor, which can result in a loss of control or even a crash. Existing solutions, such as Active Fault-Tolerant Controllers (AFTC) and tilt-rotor mechanisms, offer stabilization but often fail to achieve zero-attitude static hover, a critical requirement for many real-time applications. These limitations also adversely affect maneuverability, rendering the hexacopter unable to complete its mission. To overcome these challenges, HEXmorph is proposed as a novel geometric morphing solution that enables the hexacopter to sweep its arms and redistribute available thrust around its Center of Mass (CoM). This approach restores control during rotor failure while maintaining zero-attitude hover and full maneuverability, allowing the mission to continue. A feed-forward neural network is employed to estimate the optimal arm sweeping angles based on CoM analysis for thrust redistribution. A global event-triggered sliding mode controller manages the arm motion and locks them in place as needed. For flight stabilization, a Modified Nonsingular Terminal Sliding Mode Controller (MNTSMC) is integrated into the lower-layer flight stack. Real-time implementation of HEXmorph confirms its ability to achieve a zero-attitude static hover and maintain control over four degrees of motion, even during single or adjacent rotor failures. This project represents a transformative step in improving the fault tolerance and mission continuity of hexacopters in adverse conditions.