Forestry Drone: Flight Under Tree Canopy

Author: Bernardo Martinez Rocamora Junior

About one third of the land mass of the Earth is covered by forests, which not only are the habitats of diversity of animal species, but also are the source of several important raw materials to human activity, such as timber, fuel wood, and cellulose pulp. According to the WV Division of Forestry, West Virginia is the third most forested state in the United States and its forests cover 78 percent of the state’s 15.4 millions acres of land. Almost all of the forested land in the state is classified as commercial forestland and the economic impact of the wood products industry in West Virginia exceeds four billion dollars annually [1]. Over the course of history, mechanization and automation boosted the productivity of the Forestry industry by several orders of magnitude. But there is still a lot of space for innovation and use of the latest Robotics developments in this important industry.

Governments, entrepreneurs and academia will have to make a substantial effort to meet the increasing demand caused by population growth and higher standards of living; we will need another wave of disruptive innovations in Forestry to guarantee an even higher productivity without jeopardizing the environment. Even small contributions of Robotics can have a great impact in current productivity.

A major concern when planning the flight operations was not crashing the insect disperser drones. As it was difficult to obtain all information about the flight environment beforehand, the safety risks greatly constrained the navigation plans. The possibility of flying below 15 meters of the height over the top of the trees, or even through the forest, was unthinkable due to the risk of collision with the vegetation. And, regardless of these careful operation plans, drone crashes happened, bringing financial losses and decreasing operational efficiency. This research deals with one the bottlenecks of Aerial Robotics: the complexity of motion in cluttered environments.

Although the applications for the proposed project are not exclusive to the Forestry context, the problem can be defined as a mission in which a global plan, illustrated in Figure 1a, requires the robot to traverse the forest as fast as possible following a certain trajectory. Throughout its execution, this plan needs to be changed locally according to the obstacles detected by the robot sensors (Figure 1b) and, therefore, the operation of robots in this context requires a design with both navigation and obstacle avoidance strategies. Both requirements bring challenges on their own. Navigation requires accurate positioning of the robot, but Global Positioning System (GPS) does not provide a good solution below tree canopies due to poor satellite signal reception under the high density foliage [2]. In a forest environment, the measured errors can be as big as 50 meter, making GPS impractical for precise localization and navigation. Therefore, sensors based on vision, ranging, haptics and other perception capabilities need to be introduced for accurate flights.

Figure 1- a) Navigation plans.

Figure 1-b) Obstacle avoidance along flight.

Description: Navigation of aerial robot in forest environment. A mission is defined and a global plan is drawn (a), the robot autonomously makes local trajectory deviations to avoid trees and maintain its safety (b).

At the same time, these environments can have unexpected obstacles with a random distribution, requiring intelligent obstacle avoidance techniques using sensors capable of understanding the surroundings of the robot.But these sensors can only provide meaningful information at a maximum rate and this, combined with the robot dynamics, limits the corrections of the robot trajectory at high speeds. The statement of this problem is inspired by the flight of birds in forests and recent studies have focused on calculating the maximum speed theoretically achievable [3, 4], but there are a lot of development to be done regarding the practical implementation. My short-term goal is to propose strategies for high speed traverse in these cluttered environments using limited perception for both motion planning and obstacle avoidance, building up in recent developments that started working with these practical limitations [5, 6].

The research plan is divided in four parts: first, a literature review on high-speed navigation without the use of GPS through a randomly-generated obstacle field when only the statistics of the obstacle generating process are known a priori. Second, a method to solve the problem will be proposed, implemented and compared to other methods. Our perspective is that ranging sensors, e.g. LIDARs, can be coupled to the motion planning algorithm directly. For this, a deterministic and offline computed graph will be used and pruned online according to the sensor measurements. This pruning is intended to be fast by design and accelerated using parallel computing on GPUs (graphics processing units). Challenges include data structure, the pruning algorithm, how to connect consecutive frames and graphs and how to simultaneously map and create an inventory of the forest trees. Third, the integration of these materials in a application development environment that our laboratory got access, the AWS Robomaker [7]. This is a service provided by Amazon that extends the Robot Operating System (ROS) framework with cloud services. It also provides a robotics simulation service, which speeds-up application testing. A fourth and final step of the research would be conducing experiments by implementing the navigation solution in a drone and flying in a real forest. This will be executed in cooperation with researchers and faculty of the Davis College of Agriculture, Natural Resources and Design.

References

[1] Randall A. Childs. West Virginia’s forests. Growing West Virginia’s future.https://www.wvforestry.com/pdf/Economic%20Impact%20Study.pdf. Accessed: 2020-05-05.

[2] Bruno Siciliano and Oussama Khatib. Springer Handbook of Robotics. Springer, 2016.

[3] Sertac Karaman and Emilio Frazzoli. High-speed flight in an ergodic forest. In2012 IEEE International Conference on Robotics and Automation, pages 2899–2906. IEEE, 2012.

[4] Sanjiban Choudhury, Sebastian Scherer, and J Andrew Bagnell. Theoretical limits of speed and resolution forkinodynamic planning in a poisson forest. Robotics: Science and System Proceedings, 2015.

[5] Davide Falanga, Suseong Kim, and Davide Scaramuzza. How fast is too fast? the role of perception latency in high-speed sense and avoid. IEEE Robotics and Automation Letters, 4(2):1884–1891, 2019.

[6] Antonio CB Chiella, Henrique N Machado, Bruno OS Teixeira, and Guilherme AS Pereira. Gnss/lidar-based navigation of an aerial robot in sparse forests. Sensors, 19(19):4061, 2019.

[7] AWS RoboMaker Simulate and deploy robotic applications at cloud scale. https://aws.amazon.com/robomaker/. Accessed: 2020-05-05.