Flying Pen

Real-Time Contact-Aware Control for Aerial Manipulator via Normal Vector Estimation on Unstructured Surfaces

Project Period: Jan. 2025 – Present

Manuscript in preparation — IEEE Robotics and Automation Letters (RA-L).

🔹 Overview

• Contact Force Estimation: Estimates the interaction wrench from momentum-observer residuals while adapting thrust effectiveness online.
• Surface Normal Estimation: Estimates the local surface normal from the contact wrench and contact-consistent motion information.
• Surface Swiping Control: Modulates tangential velocity in real time based on local surface-normal variation to maintain stable contact over unstructured surfaces.

🎥 Preliminary Experiment

Early-Stage Demonstration

flyingpen_exp.online-video-cutter.com.mp4

Early-stage demonstration using an initial version of the Flying Pen framework.

Thrust Effectiveness Correction

2026-09-20.040925.mp4

Comparison between the raw momentum observer (MOB) and the η_T-updated MOB, showing improved alignment of the estimated force direction with the wall normal.

🔹 Repository Structure

• firmware — FLYINGPEN-FIRMWARE, modified Crazyflie firmware for real-world experiments, contact-force estimation, thrust-effectiveness adaptation, and contact-aware control.
• ros2_interface — FLYINGPEN_ROS2-INTERFACE, ROS 2 interface for commanding, experiment control, and flight-data logging.
• MuJoCo Simulation — MuJoCo-based environment for simulation and validation of the Flying Pen algorithms.

🔹 Previous Work

The initial surface-normal-estimation study was presented as a poster at KRoC 2025.

🔗 KRoC 2025 — Normal Estimation