Autonomous Tunnel Mapping

Project Title

Autonomous Tunnel Mapping

Project Area of Specialization RoboticsProject Summary

In this project, we aim to train a UAV for an autonomous flight inside a tunnel of unknown structure and geometry. Most of the underground tunnels are inaccessible to human and survey technologies. Therefore, we are using a customized UAV to span through the tunnel and collect the data from on-board proximity sensors (SONAR), which will be then used to make a 3D model of that tunnel. Our project aims to overcome a number of engineering challenges in making our UAV capable of autonomous flight in those areas where there is no GPS facility. For this purpose, we have used the concept of machine learning, AI, embedded systems, deep neural networks, driverless systems and flight simulators.

Project Objectives

This project has been designed to provide benefits to mainly mining community. The UAV will be able to enter underground tunnels that are inaccessible to humans and other mining equipment and will provide its 3D structure. Primary example of its use is in stope mining, shaft mining and mining of ores. The mining tunnels can sometime be hazardous to human life due to unstable roofs and walls of the mines, potential instability, unknown terrain unsuitable for ground vehicles, unknown depths of the tunnel, presence of certain harmful chemicals, and due to natural disasters (e.g. earthquake). So the UAV would be used to overcome these challenges. It will be helpful in determining the solutions for the design problem related to the tunnels and provides better planning for the further development of mines (Azhari, 2017). It will be also helpful in abolishing tunnel warfare between the miners (because Government authorities do not know the current position of the miners, so it is used to keep check and balance on them). It will also ensure fair mining practices (Government authorities can provide fair chances to the contractors and ensures that the rules and regulations imposed by the Government are being followed by the contractors).

Project Implementation Method

The Autonomous tunnel mapping is divided into four main parts:
1. Assembling of UAV
2. Algorithm Design
3. Training of Algorithm
4. Tunnel Mapping

Assembling of UAV

We used the DJI 450 frame and assembled it according to the instructions in manual. The flight controller required the calibration before making a flight. The software used by the team to calibrate and configure was Mission Planner. That can be used to calibrate the accelerometer, magnetometer and other sensors to make a stable flight of UAV. That software is also used to calibrate the RC receiver and transmitter with the Pixhawk flight controller. That calibration is then be used by the RC transmitter to transmit the values of five output channels (Roll, Pitch, throttle, Yaw and Flight Mode). That channel values will be used by the Pixhawk to make the decision about the movement, during the flight. Attached below are some pictures of the software (Mission Planner), during the configuration of Pixhawk and RC transmitter. Mounting the sensors to the UAV was the most technical and important part of the UAV’s assembly. The team then designed a data flow diagram to obtain the data that will then be used to train the algorithm, to contribute in autonomous flight and to make the three-dimensional model of that specific area. The training data should be saved to the memory to make that functional for the training of algorithm. The training data in our case is the six input values of sonar sensors and four output channels of the RC receiver.

Algorithm Design

To make our UAV autonomous we needed a deep neural network, for training and making the decisions (channel values) in real time depending on the SONAR’s values. We have used fit net function fitting the neural network in MATLAB. Since we have used six SONARs, so the input layer consists of six neurons. There are 2 hidden layers each consisting of 32 neurons and an output layer consisting of 4 neurons to provide the four channel values, for an autonomous flight. We have used three weight matrices of the following sizes: w1 (32x6), w2 (32x32) and w3 (4x32), three bias matrix b1 (32x1), b2 (32x1) and b3 (4x1) and two activation functions Rectified Liner Unit and Sigmoid.

Training of Algorithm

The Training algorithm was to be trained from these values in such a way that the algorithm must know at run-time that at what values the UAV should move right and for what values of SONAR the UAV should move left. That can basically be achieved by the predicting the values for the output channel values. The team made 80-90 flights in different environments to collect the real time data. We believed that the more data we have, more accurate results the neural networks will give.

Tunnel Mapping

We were able to obtain the trajectory of the UAV's flight. However, in order to draw a proper 3D Model a 2D Lidar sensor is required.

Benefits of the Project

This project has been designed to provide benefits to mainly mining community. The UAV will be able to enter underground tunnels that are inaccessible to humans and other mining equipment and will provide its 3D structure.

Currently no such systems are being developed in Pakistan. However, globally there is a lot of research being done on such type of projects because it has a lot of potential. These type of products are being imported from companies like Embention, Avision Robotics, Skyspecs and Emisent. So, by using the available resources we are trying to create the domestic solution on our own.
In future, we can expand our product to cope up with the agricultural needs. The UAV can be used for crop field scanning and GPS map creation through onboard camera, to have a clear understating of crop health. Thereby allowing for more efficient decision making. It can also be used by water and irrigation department to get quantitative data of canals and water channels which can then be used for analyzing and detecting the edges of water level. It can be used by the military for automatic live surveillance programs. It can also become handful in light weight package deliveries over short distances.

Technical Details of Final Deliverable

We have used a customized UAV (quadrotor) based upon the DJIF450 frame. The flight controller used for autonomous flight was Pixhawk (PX4 2.4.8) and its flight parameters were set using Mission Planner (a platform of ArduPilot for ground station controls). The UAV has been mounted with an Arduino, a raspberry Pi with a deep neural network code and SONAR sensors that will be used for the 3D mapping because there will be no clear line-of-sight in the underground tunnels. A combination of six SONAR sensors has been used that will aid in autonomous flight of UAV and visualization of the design of tunnels.

Final Deliverable of the Project Hardware SystemType of Industry IT Technologies RoboticsSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
DJI F450 GPS Drone	Equipment	1	25000	25000
Pixhawk 4	Equipment	1	8000	8000
2D Lidar	Equipment	1	37000	37000