Quantitative Evaluation of the Exploration Strategies of a Mobile Robot

How should a mobile robot explore its environment in order to build a high-quality world model as efficiently as possible? We address this question through experimentation with a sonar-equipped mobile robot. The robot is taken to be a delivery robot, such as could be used in an office, hospital or home. Its objective is to execute efficient collision-free paths between user-specified locations. A grid-based free-space map is generated for this purpose. This map is derived from a feature-based map, built using techniques similar to those of (Leonard & Durrant-Whyte 1992). Before starting to evaluate an exploration strategy, it is vital to have a clear definition of map quality. Previous map-building research has typically judged map quality either by visual inspection or by measuring the robot’s success in achieving its goals with a completed map. Neither approach provides an objective quality measure during map construction. There is therefore a need for a quantitative measure which can be applied during exploration. We solve this problem by defining a small number of numeric measures which together predict the robot’s efficiency if it were to use its current world model to achieve its objectives. The quality of the robot’s map is measured by comparing its performance in a set of tasks using either the robot’s map or an ideal map. The set of benchmark tasks is created by selecting pairs of locations such that there is an executable path between them, according to the ideal map. For each pair of locations, an attempt is then made to plan a path between them, using the robot’s map. Counts are kept of the numbers of these paths which fall into each of three categories; “Impossible”, “Collision” or “Feasible”. A path is “Impossible” if the current map shows either that one of the endpoints is occupied or that the path is blocked. A path is a “Collision” if the current map shows the route to be possible, but in fact the planned route would cause a collision with an obstacle. A path is “Feasible” if the planned route is possible without collision. In this case, we compare the cost of executing the planned route with the Ydeal” cost from the true map. The categorisation of the paths and the ef-