Proceedings of the 32nd ISR(International Symposium on Robotics), 19-21 April 2001 Vision Guided AGV Using Distance Transform Yew Tuck Chin, Han Wang 1 , Leng Phuan Tay, Hui Wang, William Y. C. Soh School of Electrical and Electronic Engineering Nanyang Technological University Nanyang Avenue, Singapore 639798 Gintic Institute of Manufacturing Technology 71 Nanyang Drive, Singapore 638075 E-mail : hw@ntu.edu.sg Abstract This paper presents an approach for outdoor robot navigation using Distance Transform Methodology (DT). DT has been used in path planning for indoor robot research and can also be used to perform obstacle avoidance simultaneously. However, when it comes to outdoor robot navigation, the operating environment becomes too large and DT becomes inefficient when performing both the tasks of path planning and obstacle avoidance. Usually both tasks have to be decoupled and DT is normally only used for path planning. The newly proposed DT methodology solves this problem by optimising the DT algorithm and reducing the processing area. Simulation and actual tests had also been carried out on an autonomous mobile robot to verify that the DT methodology can integrate both the tasks of path planning and obstacle avoidance and yield encouraging results in outdoor navigation. 1. Introduction Two important tasks in autonomous robot navigation are path planning and obstacle avoidance. Path planning is defined as the work of evaluation of the information received about the robot’s localisation, the structure of the operating environment and the goal location, through multi- sensor data acquisition and finally devised a strategic action path for the robot to manoeuvre to the goal location. Obstacle avoidance is the task of devising a safe path for the robot to manoeuvre through the immediate surrounding without collision with any obstacle that is present. Some researchers [1][2][3][4] preferred to decouple both tasks in their literature and others [5] find it more efficient to integrate them together. DT methodology is versatile and can be used for path planning alone or integration of path planning with obstacle avoidance. In this paper, we will propose an implementation of DT to achieve the latter case in outdoor navigation where the operation vicinity can easily stretch to several kilometre squares in size. 1 Corresponding Author The paper is organised as follows : Section 2 gives a description of the conventional DT methodology and the environmental map used by DT. The problems encountered when using DT for performing path planning and obstacle avoidance simultaneously in real time outdoor robot navigation. Section 3 presents the newly proposed DT methodology and explicates how it solves the problems mentioned in Section 2. Simulation results, comparing the performance of the newly proposed DT methodology and the conventional DT methodology are presented in Section 4. Section 5 shows an actual implementation of the proposed DT methodology in an outdoor autonomous robot. The proposed DT methodology is simultaneously used for both path planning and obstacle avoidance. Finally the conclusions and directions for future work are drawn in Section 6. 2. Conventional DT Methodology The conventional DT methodology in space/time proposed by Jarvis [5][6] works in a three dimensional domain formed by a two dimensional tessellated map or grid map [7][8] and time, providing a basis for optimal collision-free path finding. The methodology is simple but sufficient for navigation in indoor environment where the operation area is fixed and relatively small as compared to the case in outdoor environment. The methodology is shown below: (i) The destination cell in the tessellated map is given a distance propagation cost of zero for all time instant during distance transform. (ii) All obstacle occupied cells and boundary cells are assigned with an infinity distance propagation cost (eg. 9999) for all time instant during distance transform. (iii) All unoccupied cells in the tessellated map except the destination cell are assigned with a large distance propagation cost (eg. 5000) initially. (iv) Evaluation of the distance propagation cost for every cells in the map is done in a systematic manner from top to bottom and left to right, for a period of time T,