•Developed a modular and distributed sensing system for assisting the GPS navigation.•Designed, developed, and validated multi-channel IR sensors with CANBUS communication.•Designed a fuzzy knowledge-based controller, and validated it in simulation and actual field.•Tested the functionality of the proposed solution under harsh field conditions. Autonomous navigation of mobile robots inside unstructured agricultural fields proposes serious challenges due to the extreme variations in high-density bushes, the presence of random obstacles, and the inaccuracies in the GPS and IMU measurements. Advanced perception solutions are therefore required to assist the existing GPS-based navigation and to improve the reliability of the operation. This paper reports on the development and evaluation of a modular and scalable sensing system to assist the autonomous navigation of an agricultural mobile robot by providing it with collision avoidance capabilities. The robot benefited from a four-wheel steering mechanism that could be driven remotely via a 2.4 GHz wireless transmitter and could be programmed using the Robot Operating System (ROS) to follow waypoints. Multiple arrays of Time-of-Flight and infrared sensors with independent processing units were installed on the left, right, and front of the robot to enable a distributed control system. Communication between the sensor modules was realized via a CAN network. The collision avoidance system then exchanged messages with the robot computer over Ethernet using ROS on multiple machines scheme. A virtual model of the robot with an exact sensing setup was replicated in a robotic simulator to accelerate experimenting with different control algorithms and to optimize the sensors’ functionality. The simulation scenes and dynamic models were then improved by manually driving the robot in a real berry field for collecting sensor and steering data. Results from the simulation showed that the robot was able to autonomously navigate in different tracks and stabilize itself in the presence of random obstacles using a fuzzy knowledge-based algorithm. Preliminary field tests suggested that the Exponential filter was necessary to be implemented on each sensor for removing noise and outliers. The proposed approach created a flexible framework for exchanging data between each of the sensor ECUs and preventing the robot from colliding with random obstacles in front, left, and right. The study confirmed the functionality of the affordable sensing system and control architecture and can be suggested as an alternative solution for the high-end 3D LiDARs and the complex simultaneous localization and mapping methods.