Autonomous Navigation of 4WIS4WID Agricultural Field Mobile Robot using Deep Reinforcement Learning

Tom Baby, Mahendra Kumar Gohil, Bishakh Bhattacharya · 2024

Read Autonomous Navigation of 4WIS4WID Agricultural Field Mobile Robot using Deep Reinforcement Learning by Tom Baby, Mahendra Kumar Gohil, Bishakh Bhattacharya (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-18865.txt.

Introduction

Farm robots of the future must thread between crop rows to spray and harvest, dodging odd obstacles in tight, open-air spaces. Baby, Gohil, and Bhattacharya teach a four-wheel independent steering and driving robot to navigate with deep reinforcement learning. They first compress the many steering layouts, symmetric steering, zero turn, and a sideways-crab mode among them, into just two control variables.

In simulation, the learner tracked an irregularly shaped crop row, then hopped between rows with waypoint guidance for point-to-point runs. The authors close by racing several continuous-action learning algorithms against each other. The work maps a path toward self-driving field machines for the Agriculture 4.0 era.

Catalog record

CREATORTom Baby, Mahendra Kumar Gohil, Bishakh Bhattacharya
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-18865

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Autonomous Navigation of 4WIS4WID Agricultural Field Mobile Robot using Deep Reinforcement Learning Tom Baby, Mahendra Kumar Gohil, Bishakh Bhattacharya Abstract In the futuristic agricultural fields compatible with Agriculture 4.0, robots are envisaged to navigate through crops to perform functions like pesticide spraying and fruit harvesting, which are complex tasks due to factors such as non-geometric internal obstacles, space constraints, and outdoor conditions. In this paper, we attempt to employ Deep Reinforcement Learning (DRL) to solve the problem of 4WIS4WID mobile robot navigation in a structured, automated agricultural field. This paper consists of three sections: parameterization of four-wheel steering configurations, crop row tracking using DRL, and autonomous navigation of 4WIS4WID mobile robot using DRL through multiple crop rows. We show how to parametrize various configurations of four-wheel steering to two variables. This includes symmetric four-wheel steering, zero-turn, and an additional steering configuration that allows the 4WIS4WID mobile robot to move laterally. Using DRL, we also followed an irregularly shaped crop row with symmetric four-wheel steering. In the multiple crop row simulation environment, with the help of waypoints, we effectively performed point-to-point navigation. Finally, a comparative analysis of various DRL algorithms that use continuous actions was carried out. I Introduction The global population is expanding at a rapid rate along with the demand for sustenance. FAO ( FAO ) predicts that the global population will reach 9.7 billion by 2050 [ 1 ] . To sustain this population, food production must be considerably enhanced. Traditional farming practices and the ever-decreasing availability of farm labour render manual agriculture economically untenable and inefficient. Also, due to severe resource constraints, there is an immediate need to address issues such as the indiscriminate use of agrochemicals, energy conservation, environmental contamination management, and the effects of global warming. Improving the quantity and quality of agricultural products necessitates research into the development of intelligent, self-sufficient machinery for agricultural tasks. Field Robot-based automation in agriculture has significant socioeconomic effects on consumers, as it reduces food production costs. Agricultural automation can increase productivity, enhance product quality and resource-use efficiency, reduce human drudgery and labour shortages, and improve environmental sustainability. However, manoeuvring agricultural field mobile robots through crop rows and avoiding sensitive obstacles like tender plants under outdoor conditions remains highly challenging today. This paper aims to present an efficient navigation strategy for a 4WIS4WID ( 4WIS4WID ) mobile robot for agricultural fields where crops are planted in an organized manner. The contribution includes parameterization of some of the configurations possible in 4WS ( 4WS ). This work also incorporates implementing deep reinforcement learning as a solution for crop row tracking and navigation across multiple crop rows. II RELATED WORKS A 4WIS4WID mobile robot has four steering actuators without mechanical links and four wheels that can operate independently. It is capable of movements comparable to those of omnidirectional mobile robots. In contrast to omnidirectional robots, however, the 4WIS4WID robot requires wheel orientation changes and cannot move in any direction immediately [ 2 ] . Overall, the 4WIS4WID mobile robot incorporates the benefits of various steering configurations, providing enhanced manoeuvrability and the ability to perform a variety of motions. Vision-based navigation systems are likely the most cost-effective and reliable source of information when compared to all other sensors developed to date [ 3 ] . Hough transform has significant anti-interference ability and robustness and is less affected by noise in agricultural environments [ 4 ] . However, the application of the conventional Hough transform in real-time systems is severely constrained due to the extensive computation required. The OpenCV Probabilistic Hough Transform implementation [ 5 ] demonstrates that the proposed algorithm minimizes the computation required to detect lines by exploiting the difference in the fraction of votes required to detect lines. Precision agriculture jobs often require point-to-point planning. In a complicated context, coverage path planning is not always viable because it is computationally demanding and is primarily utilized in agricultural field machines [ 6 ] . In Cell Decomposition navigation with D* Lite, the path cannot be precisely followed by a robot with medium-high computation [ 7 ] . For large-dimensional terrain, navigation utilizing the A* star algorithm is memory-intensive [ 8 ] . Current SLAM methods can only be applied precisely in a simplified environment [ 9 ] . High computational time is associated with reactive path planning approaches. Existing reactive path planning algorithms have not yet attained the required level of robustness and dependability [ 10 ] . The methods as mentioned earlier use a map of the environment to plot a course for the robot to follow as it travels to its destination. However, the 2D mapping methods are only suitable for indoor and flat terrain because it is very challenging to create a precise model of a dynamic external environment [ 11 ] . Traditional motion planning approaches are frequently inadequate for completing jobs in complex outdoor settings. Global path planning for robots with RL shows better performance in terms of path length and path smoothness compared to RRT ( RRT ) and BFS ( BFS ) [ 12 ] . DRL ( DRL ) has also been used to handle the challenge of motion planning in a high-dimensional environment, and significant progress has been made in this area. In DRL , we don’t have to annotate vast amounts of data. DRL implementation challenges include creating an effective reward function. However, it still has the problem of poor generalization to previously unseen facts [ 13 ] , which can be tackled using domain randomization. DRL was applied to the problem of robot navigation in environments with unknown, rough terrain and varying environment size and traversability [ 14 ] . The classical methods had the lowest success rate in this work. The DRL method naturally accommodates robot-terrain interactions during the learning process due to its end-to-end approach [ 15 ] . Autonomously navigating through crop rows without maintaining any maps using a skid-steered wheeled mobile robot is successfully carried out in [ 16 ] . However, this navigation scheme implies that the robot’s starting position is in one of the field’s corners and does not account for point-to-point navigation. Most current research has concentrated on using easily accessible ROS models for skid-steered or differential drive robots in path-planning investigations. Nevertheless, these studies lack consideration of both kinematic and dynamic constraints. In addition, there is a considerable gap of research on three-dimensional path planning and robots navigating through internal non-geometric obstacles. Furthermore, there is a critical gap in the current research landscape as there are few studies on the 4WIS4WID configuration combined with Deep Reinforcement Learning (DRL). Even though DRL represents a state-of-the-art technique for autonomous navigation, its use in intricate robot configurations and three-dimensional environments has not received much attention. Given these substantial gaps in the existing literature, the primary purpose of this paper is to delve into these critical research areas. It aims to contribute by exploring and advancing navigation strategies that specifically address kinematic constraints, three-dimensional environments, navigation through internal non-geometric obstacles with the help of a camera, and the integration of DRL with the 4WIS4WID robot configurations. Through this, the paper endeavours to not only fill these critical research gaps but also enhance the state-of-the-art in autonomous

Read the complete text