Abstract
This paper introduces a comprehensive framework for an Omnidirectional Aerial Manipulator (OAM) designed to execute high-precision tasks at arbitrary 3D orientations. Conventional aerial manipulators are often limited by underactuation and mathematical singularities at extreme angles. We overcome these challenges through a novel geometric Robust Integral of the Tanh of the Error (gRITE) controller and a two-step whole-body motion planner. Experimental results demonstrate centimeter-level precision and the ability to operate in complex, cluttered environments while maintaining stability at poses previously considered impossible, such as 90° pitch or upside-down hovering.
Introduction
Aerial manipulation has traditionally been constrained by the underactuated nature of standard multirotors. These platforms must tilt to translate, preventing them from maintaining arbitrary poses in $SE(3)$. Furthermore, the dynamic coupling between a robotic arm and a flying base often leads to instability during manipulation tasks.
The core contribution of this work is a system that treats the flying base as an additional set of degrees of freedom (DoF), effectively adopting a "base as a joint" philosophy. This allows the robot to reach occluded or difficult-to-access areas by tilting the entire vehicle. To achieve this, we address three primary hurdles: dynamic coupling, mathematical singularities (e.g., gimbal lock), and the computational complexity of real-time whole-body planning.
Methods
The proposed framework integrates a robust control law with a hierarchical motion planning strategy.
System Architecture
The system architecture consists of an offline trajectory generator for the end-effector and an online whole-body trajectory generator that coordinates the base and arm movements.
The gRITE Controller
For the flying base, we developed the geometric Robust Integral of the Tanh of the Error (gRITE) controller. Unlike traditional Euler-angle-based controllers, gRITE operates directly on the rotation group $SO(3)$, eliminating mathematical singularities. It utilizes a smooth tanh function within its integral term to provide robust disturbance rejection without the high-frequency chattering common in Sliding Mode Control. This ensures smooth motor commands and hardware longevity.
Two-Step Whole-Body Planner
To enable real-time operation, the motion planning is split into two phases:
- Offline Phase: A jerk-minimized trajectory is computed for the end-effector to reach its target.
- Online Phase: A Nonlinear Model Predictive Control (NMPC) solver operates at frequencies above 10Hz. It dynamically adjusts the base's 6D pose and the arm's joint angles to follow the end-effector path while ensuring stability, avoiding self-collisions, and navigating around obstacles. Obstacles are modeled as composites of ellipsoids to facilitate smooth, differentiable optimization.
Experimental Results
The system was validated using a custom OAM with a 6-rotor tilt-rotor base and a 4-DoF robotic arm. Key findings include:
- High Precision: The system achieved a position error of less than 2 cm and an orientation error of less than 3° across various manipulation tasks.
- Extreme Maneuverability: The OAM successfully performed grasping and pulling tasks while pitched at 90° and nearly 180° (upside down).
- Robustness: In comparative tests, the gRITE controller outperformed standard geometric PID and Disturbance Observers (DOB), particularly during rapid arm movements where dynamic coupling is most significant.
- Real-time Efficiency: The online planner consistently solved within 20-30 ms, allowing for immediate reactions to environmental disturbances.
Conclusion
This work demonstrates a robust and flexible framework for omnidirectional aerial manipulation. By combining geometric control with efficient whole-body planning, we enable drones to perform complex physical interactions in cluttered spaces at any orientation. Future research will focus on integrating real-time vision for dynamic obstacle avoidance and optimizing configurations for increased payload capacity.
Summary updated