💻 computer science

Omnidirectional Dual-Arm Aerial Manipulator with Proprioceptive Contact Localization for Landing on Slanted Roofs

This paper proposes a novel dual-arm unmanned aerial manipulator that uses a momentum-based torque observer to blindly estimate the inclination of slanted rooftops through physical contact, enabling robust landings on surfaces tilted up to 30.5 degrees.

Original authors: Martijn B. J. Brummelhuis, Nathan F. Lepora, Salua Hamaza

Published 2026-02-12

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Martijn B. J. Brummelhuis, Nathan F. Lepora, Salua Hamaza

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a professional delivery driver, but instead of a van, you are a drone. Most drones are like standard cars: they work great on flat, paved roads, but if you ask them to park on a steep, snowy mountain slope, they’ll likely slide, tip over, or crash.

This paper introduces a "Swiss Army Knife" drone designed to solve that exact problem. Here is the breakdown of how it works, using everyday ideas.

1. The Hardware: The "Spider-Drone"

Most drones have fixed legs or just a flat belly. If they land on a slant, they tip. The researchers built a drone with two robotic arms that can move in every direction (omnidirectional).

The Analogy: Think of a standard drone like a person trying to stand on a tilted surfboard—it’s nearly impossible to stay balanced. This new drone is more like a mountain climber with two agile arms. No matter how the ground is tilted, the drone can reach out, find a grip, and use its arms to "prop" itself up, keeping its main body level and stable.

2. The "Blind" Sense of Touch: Proprioception

Usually, drones use cameras or lasers (like LiDAR) to "see" the ground. But what if it’s foggy, dark, or raining? Cameras get blinded, and lasers can get confused by shiny or weirdly textured surfaces.

The researchers decided to give the drone a sense of "proprioception"—which is the same sense you use to know where your hand is even when your eyes are closed.

The Analogy: Imagine you are walking through a dark room. Instead of turning on a flashlight (which might give away your position or be blocked by fog), you reach out with a walking stick. The moment the stick hits a chair, you feel the "thump" through the stick and instantly know, "Aha! There is an object there, and it's at this specific height."

The drone does this with its arms. It doesn't "see" the roof; it "feels" it. It uses a mathematical trick (a "momentum-based observer") to sense the tiny vibrations and resistance in its own motors the moment an arm touches a surface.

3. The Strategy: The "Probing" Dance

The drone doesn't just dive onto the roof. It follows a smart, cautious routine:

The Probe: It hovers above the roof and gently reaches down with one arm, like a person testing the depth of a pool with their toe.
The Mapping: Once it feels a "thump," it marks that spot. Then it reaches with the second arm to find another spot.
The Math: With those two points, it calculates the angle of the slope (like drawing a line between two dots on a piece of paper).
The Landing: It adjusts its arms to match that angle and gently sets itself down, using the arms as "stabilizing legs" to ensure it doesn't tip over.

The Results: Did it work?

The researchers tested this on slopes as steep as 30.5 degrees (which is quite a sharp tilt!).

The drone was incredibly accurate. Even though it was "blind" (not using cameras to measure the slope), it guessed the angle of the roof with an average error of only about 2.8 degrees. In the world of flying robots, that is like being able to thread a needle while riding a rollercoaster.

Why does this matter?

This technology paves the way for drones to work in much tougher environments—like inspecting steep rooftops in a storm, delivering packages to uneven mountain terrain, or navigating through dark, cluttered urban spaces where cameras fail. It turns a fragile flying machine into a rugged, tactile explorer.

Technical Summary: Omnidirectional Dual-Arm Aerial Manipulator with Proprioceptive Contact Localization

1. Problem Statement

Operating Unmanned Aerial Vehicles (UAVs) in urban environments requires the ability to land on rooftops, which are frequently slanted or irregular. Conventional sensing methods for detecting surface inclination—such as vision-based (cameras) or acoustic sensors—are often unreliable due to environmental factors like poor lighting, weather conditions, or surface material properties. Furthermore, standard quadrotors struggle to align their thrust vectors with inclined surfaces, making "hard landings" risky for both the drone and the structure. While adaptive landing gear or tethered approaches exist, they often lack the versatility required for multi-purpose aerial robots.

2. Methodology

The researchers propose a shift from exteroceptive sensing (external perception) to proprioceptive sensing (internal physical feedback) to infer environment geometry through direct contact.

Morphology & Design: The authors developed a novel Unmanned Aerial Manipulator (UAM) featuring a dual-arm configuration. Unlike traditional designs where arms are fixed, this UAM uses a pivot joint mechanism (inspired by planetary gears) that allows the arms to rotate 360° around the vehicle's body, providing an omnidirectional 3D workspace. The arms are designed with low inertia (using drive belts to move servos closer to the CoM) to minimize reaction torques on the drone body.
Contact Detection & Localization: Since high-gear-ratio servos lack force transparency, the system uses a momentum-based torque observer. By monitoring the residual of the change in angular momentum (derived from IMU gyro data), the system can estimate external torques ( $\hat{\tau}_b$ $\overset{τ}{^}_{b}$ ) without dedicated force sensors.
- Localization: Once contact is detected, the system compares the observed torque with a "virtual torque" generated by the manipulator's kinematics to identify which arm made contact and at what point.
Landing Pipeline:
1. Probing: The UAM approaches the surface and uses its arms to "feel" for contact.
2. Slope Estimation: Using the contact points, the system calculates the surface normal and the inclination angle ( $\alpha$ ).
3. Planning: The UAM reorients its heading and adjusts the manipulator arm positions so the slope between the end-effectors and the landing gear matches the detected surface.
4. Execution: The vehicle descends to land with a level body orientation.

3. Key Contributions

Novel Hardware Morphology: A dual-arm UAM with an omnidirectional workspace that decouples manipulation from the drone's orientation.
Sensorless Proprioceptive Pipeline: A method for contact detection and localization that relies on momentum observers rather than expensive or weather-dependent external sensors.
Integrated Landing Strategy: A complete autonomous workflow from blind surface probing to slope-matched landing.

4. Results

The approach was validated through nine flight experiments across three different surface inclinations (11.3°, 20.6°, and 30.5°).

Robustness: All nine landings were successful, demonstrating that the UAM could effectively estimate the slope and adjust its configuration for a stable touchdown.
Accuracy: The system achieved an average surface inclination estimation error of 2.87°.
Performance: Even at steep angles of up to 30.5°, the momentum-based observer successfully distinguished between intentional maneuvers and actual physical contact.

5. Significance

This work represents a significant step toward versatile, autonomous aerial robotics. By moving away from reliance on vision and light-dependent sensors, the proposed UAM is more capable of operating in "blind" or degraded visual environments (e.g., fog, night, or dust). The omnidirectional workspace and dual-arm design transform the limbs from simple landing gear into a "Swiss Army knife" capable of complex manipulation, perching, and landing, which is essential for future urban logistics and search-and-rescue missions.