Development of a semi-autonomous framework for NDT inspection with a tilting aerial platform

TL;DR

This work tackles the challenge of performing non-destructive testing (NDT) with a semi-autonomous aerial manipulator by integrating an omnidirectional tilting UAV with a 5-DoF robotic arm. A parallel force-impedance control framework enables smooth transition from contact-free motion to contact-based interaction and a pushing phase to create a vacuum for echometer thickness measurements, with a decoupled flight/arm architecture and an impedance filter to maintain stability. The hardware platform features a coaxial tilting octa-rotor drone, a lightweight arm equipped with a 6-axis force/torque sensor, a configurable echometer probe, a battery carriage for balance, and a custom PX4 flight control stack. Experimental validation in indoor, GPS-denied environments demonstrates repeatable measurements and robust interaction, while outdoor deployment is identified as an area for future improvement, including autonomous operation and enhanced perception with lidar or GPS. The proposed approach advances NDT capabilities by delivering dexterity and precision akin to on-site operators, from remote locations.

Abstract

This letter investigates the problem of controlling an aerial manipulator, composed of an omnidirectional tilting drone equipped with a five-degrees-of-freedom robotic arm. The robot has to interact with the environment to inspect structures and perform non-destructive measurements. A parallel force-impedance control technique is developed to establish contact with the designed surface with a desired force profile. During the interaction, a pushing phase is required to create a vacuum between the surface and the echometer sensor mounted at the end-effector, to measure the thickness of the interaction surface. Repetitive measures are performed to show the repeatability of the algorithm.

Figures (7)

• Figure 1: Proposed coaxial tilting octa-rotor equipped with a robotic arm. Snapshot in laboratory (a) and during task execution (b).
• Figure 2: Kinematic structure of the five DoF arm and moving battery carriage.
• Figure 3: Overall manipulator control architecture. The arm control inputs are computed by double integrating the auxiliary input. The matrices $S_f$ and $S_v$ select the Cartesian-space axis along which performs an impedance control or force control.
• Figure 4: Experimental results: drone's reference and actual linear position along the three Cartesian axes. The UAM counter-reacts to the external disturbance during the interaction preserving its pose by tilting the propellers.
• Figure 5: Experimental results: arm's joints state feedback evolution. The impedance controller adjusts the joint angular position to be compliant w.r.t. the interaction surface.