Lidar Based Intelligent Obstacle Avoidance System for Autonomous Ground Vehicles
P. Shunmuga Perumal1, M. Sujasree2, K. Siddhardha3, K. Gokul4

1P. Shunmuga Perumal*, Automotive Research Centre, Vellore Institute of Technology, Vellore, India.
2M. Sujasree, School of Advanced Sciences, Vellore Institute of Technology, Vellore, India.
3K. Siddhardha, Department of Aerospace Engineering, IIT Madras, India.
4K. Gokul , Samsung Research and Development Institute, Bangalore, India.
Manuscript received on March 16, 2020. | Revised Manuscript received on March 24, 2020. | Manuscript published on March 30, 2020. | PP: 2466-2474 | Volume-8 Issue-6, March 2020. | Retrieval Number: F8029038620/2020©BEIESP | DOI: 10.35940/ijrte.F8029.038620

Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (

Abstract: Autonomous ground vehicles (AGVs) started occupying our day-to-day life. AGVs can be programmed to be smart with the current technological advancements. In doing so, we can apply them to assist humans in many aspects like reducing road accidents, enabling us to use cars without driving knowledge, autonomous patrolling in dangerous zones, and autonomous farming. For AGVs to operate at this level of automation, it must be equipped with sensory perception devices to be aware of its surroundings, and also, a way to perceives this data is crucial. As a first step towards this, researchers have developed a vast number of camera vision-based efficient neural network algorithms for detecting and avoiding obstacles. Unfortunately, an AGV cannot survive only with computer vision as it suffers from several effects like night driving and erroneous estimation of distance information. Camera vision and lidar vision together is suitable for AGVs to operate in all conditions like day, night, and fog. We propose a novel neural network model, which transforms the lidar sensor data into obstacle avoidance decisions, which is integrated into the hybrid vision of any AGV. Existing lidar sensor-based obstacle detection and avoidance systems like 2D collision cone approaches are not suitable for real-time applications, as they lag in providing accurate and quick responses, which leads to collisions. The proposed intelligent Field of View (FOV) mechanism replaces classical mathematical approaches, which accurately mimics the behavior of human drivers. The model quickly takes decisions with a high level of accuracy to command the AGV upon being obstructed with obstacles in the trajectory. This makes the AGV drive in obstacle rich environments without manual maneuvering autonomously.
Keywords: Lidar Sensor-Based Feed Forward Neural Network, Autonomous Ground Vehicles, Obstacle Detection And Avoidance.
Scope of the Article: Autonomous Robots.