Introduction
Standard quadcopters are inherently underactuated, meaning they cannot move laterally without tilting their entire body. This coupling of translational and rotational motion poses a significant challenge for aerial manipulation tasks, such as drilling or grinding, which require applying precise forces to a surface while maintaining stability. While adding robotic arms is a common solution, it introduces significant weight penalties and complex dynamical coupling. This paper presents a novel 5-degree-of-freedom (DoF) tilting-rotor UAV designed to overcome these limitations through a unique "H" frame mechanical design and an integrated cyber-physical system (CPS).
System Architecture
The proposed system integrates mechanical innovation with advanced control and onboard vision to enable autonomous interaction.
Mechanical and Control Design
The UAV utilizes an "H" configuration with two additional servomotors that tilt pairs of rotors. This design allows the vehicle to generate 5-DoF thrust forces and torques, effectively decoupling longitudinal propulsion from orientation. To manage environmental interaction, a dual-level cascaded control system is employed: a low-level PID controller for attitude and a high-level selective impedance controller. This architecture allows the drone to "feel" and regulate contact forces when interacting with surfaces, such as a wall or workpiece.
Cyber-Physical Integration and Vision Pipeline
To minimize latency, the vision-based tracking system is implemented entirely onboard. The pipeline uses YOLO v3 for object detection, a KLT (Kanade-Lucas-Tomasi) tracker for 2D motion, and morphological operations for pinpointing targets. To ensure real-time performance on edge hardware like the NVIDIA Jetson Nano or FPGAs, the authors applied model compression techniques, including weight pruning and quantization. This "brain-on-body" approach eliminates the transmission delays associated with off-board processing.
Key Results and Insights
The system successfully achieves 5-DoF actuation, resolving the underactuation issues of traditional quadcopters without the mechanical complexity of a fully 6-DoF design. Key performance metrics include:
- Vision Processing: The optimized YOLO architecture achieved 20 FPS on an NVIDIA Jetson TX2 and reached a blistering 314.2 FPS on a Xilinx FPGA, with seven times higher energy efficiency.
- Control Stability: The impedance control law successfully decoupled translational and rotational dynamics on the longitudinal plane, enabling stable trajectory tracking during contact.
- Efficiency: By adding only two servomotors instead of additional rotors, the design avoids "force cancellation," preserving battery life and payload capacity.
Conclusion
This research demonstrates that high-DoF aerial manipulation is achievable through a combination of simplified mechanical design and high-performance onboard computing. By moving the computation to the UAV and using impedance control to bridge the gap between flight and contact, the system provides a robust platform for autonomous industrial tasks. Future work will focus on exploring full 6-DoF omnidirectionality and testing the system's precision in more diverse real-world environmental conditions.
Summary updated