An Exploration Method for General Robotic Systems Equipped with Multiple Sensors Luigi Freda, Giuseppe Oriolo and Francesco Vecchioli Abstract— This paper presents a novel method for sensor- based exploration of unknown environments by a general robotic system equipped with multiple sensors. The method is based on the incremental generation of a configuration-space data structure called Sensor-based Exploration Tree (SET). The expansion of the SET is driven by information at the world level, where the perception process takes place. In particular, the frontiers of the explored region efficiently guide the search for informative view configurations. Different exploration strategies may be obtained by instantiating the general SET method with different sampling techniques. Two such strategies are presented and compared by simulations in non-trivial 2D and 3D worlds. A completeness analysis of SET is given in the paper. I. INTRODUCTION This paper presents a novel exploration method by which a general robotic system equipped with multiple sensors can explore an unknown environment. The method is suitable for generic robotic systems (such as fixed or mobile ma- nipulators, wheeled or legged mobile robots, flying robots), equipped with any number of range finders. In a sensor-based exploration, the robot is required to ‘cover’ the largest possible part of the world with sensory perceptions. A considerable amount of literature addresses this problem for single-body mobile robots equipped with one sensor, typically an omnidirectional laser range finder. In this context, frontier-based strategies [1]–[5] are an in- teresting class of exploration algorithms. These are based on the idea that the robot should approach the boundary between explored and unexplored areas of the environments in order to maximize the expected utility of robot motions. The problem of exploring an unknown world using a multi-body robotic system equipped with multiple sensors is more challenging. In fact, the sensing space (the world) and the planning space (the configuration space) are very differ- ent in nature: the former is a Euclidean space of dimension 2 or 3, while the latter is a manifold in general with dimension given by the number of configuration coordinates, typically 6 or more. While frontiers at the world level clearly retain their informative value, using this information to efficiently plan actions in configuration space is not straightforward. In the literature, few works exist that address the sensor- based exploration problem for articulated structures, mainly for fixed-base manipulators equipped with a single sensor, Luigi Freda, Giuseppe Oriolo and Francesco Vecchioli are with Dipartimento di Informatica e Sistemistica, Uni- versit` a di Roma “La Sapienza”, Via Ariosto 25, 00185 Roma, Italy, {freda,oriolo}@dis.uniroma1.it, francesco vecchioli@fastwebnet.it This work has been funded by the European Commission’s Sixth Frame- work Programme as part of the project PHRIENDS under grant no. 045359. e.g., see [6]–[9]. A related problem is 3D object reconstruc- tion and inspection [10]. The SET (Sensor-based Exploration Tree) method, which was originally presented in [11] for single-sensor robotic systems, is a frontier-based exploration method. The basic idea is to guide the robot so as perform a depth-first exploration of the world, progressively sensing regions that are contiguous from the viewpoint of sensor location. In this process, frontiers are used to efficiently identify informa- tive configurations. The information gathered about the free space is mapped to a configuration space roadmap which is incrementally expanded via a sampling-based procedure. The roadmap is used to select the next view configuration, which is added to the SET. In the exploration process, the robot alternates forwarding/backtracking motions on the SET, which essentially acts as an Ariadne’s thread. In this work, we present (i) an extension of the SET method to multi-sensor robotic systems (ii) a completeness analysis of the algorithm (iii) a SET implementation on non-trivial 2D and 3D worlds. In particular, we discuss how to identify which frontiers are relevant for guiding the perception of each sensor and how to assign priorities to the sensors during view planning. The paper is organized as follows. The problem setting is given in Sect. II. A general exploration method is outlined in Sect. III and the SET method is presented in Sect. IV. Simulation results in different worlds are reported and dis- cussed in Sect. VI. Some extensions of the present work are mentioned in the concluding section. II. PROBLEM SETTING The robot wakes up in a unknown world populated by obstacles. Its task is to perform an exploration, i.e. cover the largest possible part of the world with sensory percep- tions [12]. A. Robot and World Models The robot, denoted by A, is a kinematic chain of r rigid bodies (r ≥ 1) interconnected by elementary joints. This description includes: fixed-base manipulators, single-body and multiple-body mobile robots, flying robots, humanoids and mobile manipulators. The world W is a compact connected subset of IR N , with N =2, 3. It represents the physical space in which the robot moves and acquires perceptions. W contains the static obstacles O 1 , ..., O p , each a compact connected subset of W. One of these obstacles is the world boundary ∂ W which is considered as a ‘fence’. Denoting by O =  m i=1 O i the obstacle region, the free world is W free = W\O. The 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems October 11-15, 2009 St. Louis, USA 978-1-4244-3804-4/09/$25.00 ©2009 IEEE 5076