Introduction
Traditional multirotor aerial vehicles, such as standard quadcopters, are typically underactuated. This means their translational and rotational degrees of freedom are coupled; for instance, a quadcopter must tilt forward to gain horizontal velocity. This coupling limits their agility in constrained environments. In contrast, Omnidirectional Multirotor Aerial Vehicles (OMAVs) are fully actuated, possessing the ability to control their 3D position and 3D orientation independently. While this provides a significant advantage for navigating cluttered spaces, it introduces a complex trajectory planning problem in the six-degree-of-freedom (6-D) space, specifically within the Special Euclidean group $SE(3)$.
The primary challenge in 6-D planning lies in the rotation manifold $SO(3)$. Traditional representations like Euler angles suffer from gimbal lock, while quaternions require unit-norm constraints that complicate optimization. This research introduces a three-stage framework designed to generate safe, smooth, and efficient 6-D trajectories by mapping rotations into a free vector space using stereographic projection.
The Three-Stage Optimization Framework
The proposed pipeline decomposes the trajectory generation process into three distinct phases to balance global connectivity with local smoothness and safety.
1. Path Search
The process begins with a standard Rapidly-exploring Random Tree (RRT) algorithm. This stage identifies a rough, collision-free path from the start configuration to the goal. While this path is not optimized for smoothness or vehicle dynamics, it provides a topological skeleton for the subsequent stages.
2. Safe Flight Corridor (SFC) Generation
To ensure the vehicle remains clear of obstacles, the framework converts the rough path into a Safe Flight Corridor (SFC). This corridor consists of a series of intersecting convex polyhedra. These volumes represent the guaranteed "safe" airspace within which the trajectory must be contained.
3. 6-D Trajectory Optimization
The core of the research is the optimization stage, which refines the rough path into a high-quality 6-D trajectory. This stage utilizes several key innovations:
- Stereographic Projection (SP): To circumvent the mathematical constraints of $SO(3)$, the authors project the vehicle's attitude from a 4D quaternion onto a 3D vector space. This transformation allows the rotation trajectory to be treated as a curve in $R^3$, enabling the use of high-performance, unconstrained "quasi-Newton" solvers like L-BFGS.
- MINCO Representation: The trajectory is represented using the Minimum Control Effort (MINCO) class of polynomials. This allows the system to minimize "snap" (the fourth derivative of position) and total flight time simultaneously, ensuring the resulting motion is smooth and executable by the hardware.
- Whole-Body Safety: Unlike traditional planners that simplify drones as spheres or points, this framework models the OMAV as a cuboid. The optimization ensures that all eight vertices of the cuboid remain within the SFC. This "whole-body awareness" allows the drone to calculate the exact tilt required to pass through narrow apertures.
Experimental Evaluation
The framework was evaluated through both extensive simulations and real-world hardware tests.
Simulation Scenarios
The researchers tested the algorithm in three distinct virtual environments:
- Narrow Passage: A vertical gap only 0.7m wide. Given the OMAV (OmniHex) has a width of 1.1m, the vehicle was forced to tilt significantly to pass through—a maneuver impossible for underactuated drones.
- Cluttered Environment: An unstructured space requiring complex 6-D maneuvers.
- Large-Scale Random Forest: A scenario featuring 1,000 randomly distributed cuboid obstacles to test the robustness and scalability of the solver.
Hardware Implementation
Real-world experiments were conducted using the "OmniHex," a custom-built tilt-rotor hexarotor. The hardware tests confirmed that the trajectory optimization could be performed in near-real-time, with computation times ranging from 88ms to 611ms depending on the complexity of the environment and the number of obstacles.
Key Results and Insights
The results demonstrate that the Stereographic Projection (SP) method is significantly more stable and computationally efficient than traditional Euler-angle-based optimization. In large-scale simulations, the SP method showed much lower variance in convergence times, making it more predictable for real-time deployment.
Furthermore, the ability to decouple position and orientation allowed the OmniHex to navigate gaps narrower than its own diameter by adopting a "squeezing" posture. By softening kinodynamic limits (velocity and acceleration) into penalty terms within the objective function, the framework maintains high performance without the overhead of hard constraints.
Conclusion and Future Work
This research provides a robust foundation for 6-D OMAV navigation in static, cluttered environments. By transforming the rotation manifold into a flat vector space, the authors have made 6-D trajectory optimization practical for real-world robotics.
Future work will focus on extending this framework to dynamic environments containing moving obstacles. Additionally, the team aims to apply these 6-D trajectories to autonomous aerial manipulation tasks, where the OMAV's ability to exert forces in any direction while maintaining a specific orientation is a critical advantage.
Summary updated