Over the last few decades, robots have been more often in use for both industry and daily life, thanks to the breakthroughs in technology. Robots have flexibility of being capable of performing multi-task applications simultaneously. Further, they are more accurate and consistent than human, which results in increased productivity and profit.
A mobile robot is a system that is capable of autonomous locomotion around its environment. Navigation is one of the most important tasks of mobile robots. Staying operational while avoiding collisions and maintaining safety standard are priorities in mobile robots 1. To build an autonomous mobile robot, we have to build a system that can grasp environments, react to unexpected events, and plan dynamically in order to achieve the mission. Thus, the objective of the robot’s motion planning and control is to find collision-free paths between two positions in static or dynamic environments. In this context, control has different levels. This includes low-level motor control and behavior control. The latter represents many complicated tasks, such as obstacle avoidance and goal seeking 2. The basic goal of mobile robot navigation in an unknown environment is to avoid obstacles and to correct the waypoint from the new position in real-time 3.
Despite impressive development in the field of autonomous robotics in the last two decades, robot navigation is still an area of active research due to the uncertainties involved with unknown real-life environments. These uncertainties could be attributed to any of the following: (i) no prior knowledge of the environment, (ii) perceptually acquired information is usually unreliable, (iii) unpredictable complex dynamical environment, and (iv) the effect of control actions is not completely reliable 4, 5. Classical navigation approaches relied on geometric models like constructing local cost-maps, which is considered as a low-level intelligence without any perception process. There are also some unsolved challenges with regard to perception and intelligent control. The perception of the environment relies on various sensor information. Classical perception methods extract information from the raw sensor readings based on artificially designed complex features (e.g., SIFT, SURF, ORB). Most of those methods are designed to adapt to generic environments. Nevertheless, regarding unstructured dynamic environments, those methods are prone to errors. On the other hand, the control system should realize appropriate decision-making policy effectively and efficiently based on environmental perception. For example, the navigation for mobile robots with coarse precision is considered solved. However, in challenging environments or when precision is of high importance over large-scale 6, these approaches lack robustness and are unable to overcome the above-mentioned challenges. Therefore, a development of intelligent control strategies becomes a must.