Real-time localisation and path-planning in ackermann steering robot using a single RGB camera and 2D LIDAR

The rise in road accidents due to human errors and the pollution caused due to traffic has led to an increase in the development of technology required for making autonomously navigating vehicles. This has also led to development of numerous rapid prototyping platforms and techniques which allow us to test various path planning algorithms and control methods. In this paper a novel and modular experimental setup to perform localisation, obstacle detection and path planning for an Ackermann steering robot using only a single RGB camera and a LIDAR sensor has been presented. The RGB camera is mounted on the ceiling facing directly towards the ground and connected to an external laptop. For localisation, a uniquely shaped marker placed on the robot is used to find the position and the orientation of the robot in the video stream coming from the camera. This is done using color filtering followed by blob detection. The mobile robot has a LIDAR sensor and an on-board computer which is used to transfer the LIDAR data through a wireless network to the external laptop using TCP/IP protocol. The obstacle data from the robot is enhanced using potential field and combined with the position data from the overhead camera to construct a cost-map which is a detailed representation of the position of the robot and all the obstacles around it. The cost-map is used to find a path around the obstacles to reach the destination depending upon the starting position and orientation. This setup was tested using four different types of search algorithms namely Uniform Cost Search, A-star(A*), Dynamic Programming(DP) and Hybrid A* and the results have been reported. The results verify the superiority of Hybrid A* over other algorithms. An optimization technique called gradient descent is used to smooth out the path obtained