Short-Safe Compromise Path for Mobile Robot Navigation In A Dynamic Unknown Environment Sardjono Trihatmo R.A. Jarvis Intelligent Robotic Research Center Monash University, Wellington Road, Clayton Victoria 3168, Australia Sardjono@inn.bppt.go.id Ray.Jarvis@eng.monash.edu.au Abstract This paper presents a path planning method for a mobile robot in an environment that is unknown and can change. This method relies on continually sensing the local environment. Robot and personnel safety is a dominant and essential requirement for this path planning method. A new method for obtaining safe directions in a tessellated map of the environment is introduced. Linear combination of vectors is used to obtain the path that is a compromise between the safest and the shortest path. The path enables a mobile robot to reach its goal directly and safely. 1. Introduction Successful navigation of mobile robots requires planning a path that enables the robot to reach a predefined goal without collision with obstacles. Two path planning methods that are widely used in robotic research are the A* search method [Nilsson, 1971 & Jarvis, 1983] and the Distance Transform (DT) method [Jarvis, 1993]. The A* search algorithm determines the shortest path among polyhedral obstacles. The Distance Transform algorithm generates the minimum number of steps, thus the shortest path, in a cell based environmental model. Although both path-planning algorithms (A* and DT) are designed so that the robot avoids collision, there is a tendency for the robot to be driven close to obstacles. Consequently, there is a risk of collision since perfect motion control and exact data about the environment are difficult to obtain. Zelinsky introduced a safe path transform method in his work [Zelinsky, 1994]. This method, which is an extension to the DT method, generates a path that is safer than that of the basic DT method in a known environment. The randomized potential field method [Barraquand and Latombe, 1991] generates a safe path, but the path tends to stray far from the direct path to the goal. The other problem of the randomized potential field method is the existence of local minima at which the robot may get stuck. Other path planning algorithms such the randomized roadmap [Amato and Wu, 1996] and rapid- exploring random tree [La Valle and Kuffner, 1999] are powerful in a cluttered configuration space and the kinodynamic planning problem, respectively. However, both path planning algorithms do not state the safeness of the path explicitly. In a dynamic unknown environment, a global path to the goal must be recalculated every time the global environment changes or some new aspect of it is discovered as the robot moves. Otherwise, collisions easily occur when using out of date mapping data as a basis for path planning. Instead of determining a global path and the necessity to recalculate it every time the environment changes, it is useful and essential to develop a path planning method that relies on continually sensing local environments. This so called local path planning method obtains a local target in a local environment such that the mobile robot can move to this local target safely. At the local target the environmental map will be updated and the next local target found and so on until the goal is reached. It must be assumed that the local environment does not change when the robot is moving to the local target. Lipton has developed the sophisticated local next step path planning method [Lipton, 1997] for navigation amongst polyhedral obstacles which are previously unknown. The local target is obtained by using a combination of the desirability and danger function. However, there is a requirement to find a cul de sac in advance to enable the robot to avoid it. This paper presents a path planning method for a dynamic unknown environment which is tessellated. This path planning method estimates the shortest global path and obtains the safe local direction in which a mobile robot moves safely in a local environment. The 1