💻 computer science

Real-World Deployment of a 5G-Connected Edge-Controlled Aerial Robot in Industrial Subterranean Mines

This paper presents the first real-world demonstration of an autonomous aerial robot navigating an active industrial subterranean mine via a 5G Standalone network, utilizing a Kubernetes-based edge cluster to host a Model Predictive Controller that generates safe, collision-free paths from human-selected waypoints.

Original authors: Achilleas Santi Seisa, Emanuele Pagliari, Gerasimos Damigos, Elias Small, George Nikolakopoulos

Published 2026-06-04

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Achilleas Santi Seisa, Emanuele Pagliari, Gerasimos Damigos, Elias Small, George Nikolakopoulos

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 a tiny, high-tech drone trying to fly through a dark, twisting, and dangerous underground mine. Usually, this drone would need a super-brain built right into its tiny body to figure out how to fly, avoid walls, and follow a human's instructions. But that super-brain is heavy, eats up battery power, and is hard to upgrade.

This paper describes a clever workaround: They gave the drone a "remote brain" instead.

Here is how they did it, broken down into simple concepts:

1. The Setup: A Drone with a "Remote Brain"

Think of the drone as a car that has no engine or computer inside it. Instead, it has a tiny radio.

The Drone: It's a standard flying robot (a quadcopter) equipped with a laser scanner (like a high-tech flashlight that sees in 3D) to see the mine walls.
The Remote Brain: Instead of thinking for itself, the drone sends a quick text message to a powerful computer cluster located just outside the mine (called the "Edge"). This cluster is like a massive server farm sitting in a nearby building.
The Connection: They are connected by a 5G network. Think of this not just as fast internet, but as a dedicated, ultra-reliable "super-highway" for data that doesn't get jammed, even deep underground.

2. The Game of "Hot Potato" (How it Works)

The process happens in a split-second loop, like a game of hot potato played at the speed of light:

The Human: A person sitting safely outside the mine picks a destination on a map (a "waypoint").
The Drone: The drone takes a snapshot of where it is and what it sees, then shoots that data up the 5G highway to the remote brain.
The Remote Brain: The powerful computer outside calculates the perfect path to the destination, making sure the drone doesn't crash into rocks or tunnels. It's like a chess grandmaster calculating the next move.
The Return: The brain sends the instructions (turn left, go up, speed up) back down the 5G highway to the drone.
The Action: The drone executes the move immediately.

3. Why This is a Big Deal

Usually, scientists test these ideas in computer simulations or clean, quiet university labs. This paper claims to be the first time this specific setup has been tested in a real, active, industrial mine.

The Challenge: Mines are messy. The air is dusty, the tunnels are narrow, and the radio signals can be tricky.
The Result: The system worked. The drone flew smoothly, followed the human's chosen path, and avoided collisions, all while its "brain" was miles away in a different building.

4. The "Speed of Thought" (Latency)

For this to work, the "conversation" between the drone and the brain has to be incredibly fast. If the signal takes too long, the drone might crash before it gets the "stop" command.

The researchers measured the time it took for the signal to go up and come back (Round-Trip Time).
They found the delay was very small (mostly between 4 and 13 milliseconds).
The Analogy: Imagine throwing a ball to a friend and catching it back before you even finish blinking. That's how fast the system reacted. Because it was so fast, the drone stayed stable and safe.

5. The Main Takeaway

The paper proves that you don't need to pack heavy, expensive computers onto every robot. Instead, you can send the heavy thinking to a powerful computer nearby, as long as you have a fast, reliable 5G connection.

In short: They successfully flew a drone in a real mine using a "remote control" that was actually a super-computer thinking for the drone, proving that this technology is ready for the real world, not just the lab.

Technical Summary: Real-World Deployment of a 5G-Connected Edge-Controlled Aerial Robot in Industrial Subterranean Mines

Problem Statement
While the integration of 5G New Radio (NR) Standalone (SA) networks and edge computing for robotic systems has been extensively studied in simulations and controlled laboratory environments, there is a significant gap in real-world validation, particularly within hazardous industrial settings. Existing research often fails to address the specific challenges of deploying time-critical, closed-loop control systems in uncontrolled environments like active subterranean mines. Key obstacles include communication asymmetry (where downlink latency is typically optimized over uplink), the computational limitations of onboard robotic processors for running advanced algorithms like Nonlinear Model Predictive Control (NMPC), and the need for robust, low-latency bidirectional communication to ensure stability in networked control systems.

Methodology and System Architecture
The authors present a real-world implementation of a 5G-connected aerial robot (quadcopter) controlled by an edge-offloaded NMPC. The system architecture comprises three primary components:

Robotic Platform: A custom lightweight quadcopter equipped with an Ouster LiDAR for environmental perception and an onboard Intel NUC running Ubuntu 20.04 and ROS 1. To minimize payload weight, the drone utilizes a custom 5G setup featuring a Sierra Wireless EM9293 modem and four antennas for spatial coverage. Localization is handled onboard to avoid transmitting high-bandwidth point cloud data; only telemetry and command actions (kilobytes of data) are transmitted.
5G Infrastructure: The deployment utilizes an Ericsson 5G NR SA network adapted for mining conditions. This includes macro base stations for outdoor coverage (3.5 GHz, Band N78) and customized two-sector base stations with highly directive antennas for indoor tunnel coverage. The network employs multiple slices, with one specifically optimized for low-latency experimental applications.
Edge Computing Cluster: The high-level controller is offloaded to an on-site Kubernetes (k8s) cluster located in a dedicated enclosure. The cluster consists of four nodes with Intel multi-core CPUs, 590 GiB RAM, and NVIDIA GTX 2080 Ti GPUs. The NMPC controller runs as a containerized application within a dedicated k8s pod, managed via Rancher Kubernetes Engine (RKE). Communication between the robot and the edge is facilitated via User Datagram Protocol (UDP) over the 5G link.

Control Strategy and Delay Management
The system employs Nonlinear Model Predictive Control (NMPC) to generate smooth, collision-free trajectories based on waypoints selected by a human operator. The control loop is subject to round-trip time (RTT) delays, defined as $t_{rtt}(t) = t_u(t) + t_d(t) + t_p(t)$ , where $t_u$ and $t_d$ are uplink and downlink transmission delays, and $t_p$ is the edge processing time. The authors explicitly account for the asymmetry of cellular networks, where uplink delays can be higher than downlink delays, by measuring these specific delays during the experiment and incorporating them into the stability analysis.

Experimental Results
The system was validated in an active industrial mine in northern Sweden. A human operator selected waypoints, and the edge-offloaded controller successfully guided the robot through the environment.

Trajectory Tracking: The robot accurately followed the reference waypoints in x, y, and z planes, demonstrating the feasibility of the closed-loop system.
Network Performance: Measurements collected via Qosium showed:
- Uplink (UL) Delay: Ranged from 5 ms to 34 ms (drift-corrected average: 9.5 ms).
- Downlink (DL) Delay: Ranged from 3 ms to 27 ms (drift-corrected average: 6.5 ms).
- Jitter: Ranged from 0–5 ms (UL) and 0–11 ms (DL).
- Throughput: Average UL traffic was 2.1 Mbps; DL was 0.1 Mbps.
Stability: The measured delays satisfied the stability condition $t_{rtt}(t) < \tau_{max}$ , confirming that the system operates within a stable regime despite the network-induced delays.

Key Contributions
The paper claims to provide the first real-world implementation and evaluation of a time-critical, 5G-connected, edge-controlled aerial robotic platform in an uncontrolled, non-laboratory environment. The specific contributions are:

Real-World Framework: The development of a time-critical closed-loop edge-based framework using Kubernetes to offload computationally intensive control modules for industrial robotics.
5G Integration: The successful integration of 5G NR SA networks to meet real-time control requirements in a harsh subterranean setting.
Feasibility Validation: The empirical validation that such systems can operate safely and efficiently in active industrial environments, bridging the gap between academic research and industrial application.

Significance and Claims
The authors position this work as a critical step toward the broader realization of edge robotics via 5G. They argue that shifting computational loads to the edge reduces the need for heavy onboard processing, making robotic platforms more affordable, lightweight, and easier to upgrade. The study demonstrates that current 5G capabilities are sufficient to support time-critical automation in hazardous environments, offering a pathway for safer and more efficient mining operations. The paper concludes modestly, noting that while the current system is successful, future work could explore more advanced data offloading and the incorporation of network link quality parameters into mission planning to further enhance reliability.