RoBotanic: a Robot Guide for Botanical Gardens. Early steps.

Transcription

1 RoBotanic: a Robot Guide for Botanical Gardens. Early steps. Antonio Chella, Irene Macaluso, Daniele Peri, and Lorenzo Riano Department of Computer Engineering (DINFO) University of Palermo, Ed.6 viale delle Scienze, Palermo, Italy chella@unipa.it,{macaluso,peri,riano}@csai.unipa.it Abstract. RoBotanic is a brand new project we recently started to deliver a robotic guide for the Botanical Garden of the University of Palermo. This project follows on the researches of the Robotics Lab that yelded CiceRobot, the robotic guide that has been tested in the indoor environment of the Archaeological Museum of Agrigento. In this paper we describe the goal of the new project, the design challenges we are facing in a difficult and vast outdoor environment, and the early steps we are undertaking to meet the goal. Key words: Personal Robotics, Cognitive Robotics, Human Robot Interaction, Robotic Guide 1 Introduction The interaction of humans and robot is a fascinating research with a broad range of possible applications. One of the most exciting is represented by a robotic tour guide for museum. The task is considered a relevant case study (see [1]) as it involves perception, self perception, planning and human-robot interaction [4]. The Robotics Lab of University of Palermo has been pursuing a robotic architecture for robots that takes into account several suggestions from cognitive science [2], [3] trying to address the problems posed by the application. The architecture has been used in the previously started CiceRobot project (see [4]), aimed at developing a robotic guide for museums. A lot of experiences and data has so far been collected concerning with both the problem of a robot guiding a group of human visitors inside a sensitive area full of precious artifacts, and the interaction of generally enthusiastic and curious humans with the technological guide. Thus we can easily affirm that simply putting a robot in the context of a museum generates a new interest in visitors, even the ones completely unaware of the role of the robot, as we often have experienced in the pauses between experiments. Such high interest, especially in pupils, has given us the motivation to keep on pursuing the development of a robotic guide. Following this inspiration we started the RoBotanic project aimed at bringing a robotic guide to outdoor environment of the Botanical Garden of the University of Palermo. We are transferring the experience that has so far been gained in the CiceRobot

2 2 Antonio Chella, Irene Macaluso, Daniele Peri, and Lorenzo Riano project by the Robotics Lab to the RoBotanic project, although major differences between the two scenarios are quite evident. In this paper we present the goal of the RoBotanic project, the current status and the steps we are undertaking, as well as as the challenges that such a difficult outdoor environment poses us. 2 The goal The RoBotanic project is focused on developing a robot able to guide visitors of the Botanical Garden of the University of Palermo to discover in a new way the wealth of plants, history, architecture, finely crafted decorations and statues that the more than two centuries old institution contains. Obviously, we are focused on tours that are considerably simpler than those offered by the human guides of the garden. Nevertheless delivering even simple robot guided tours is a challenging task when the environment is the one we describe in the following section. Fig. 1. Map of the Botanical Garden 2.1 The environment The Botanical Garden of the University of Palermo is a research and educational institution hosting more than species of plants and with an extension of about 30 acres. The origins of the garden date back to the late 18th century as reflected by the neoclassic style of the three main buildings, the Gymnasium, the Calidarium, the Frigidarium, respectively spotted by labels 1, 2 and 3 in the map of figure 1. This building complex define the north side of the original

3 RoBotanic: a Robot Guide for Botanical Gardens. Early steps. 3 structure corresponding to the rectangular area of the System of Lynnaeus (spot 4 ) sectioned by two crossing walkways. The main walkway extends for 200 meters down south (right of the map) to the round pool ( 5 ) called Aquarium hosting the aquatic plants. The orthogonal walkway, long 100 meters, walked towards west leads to the Giardino d Inverno ( Winter garden ) (fig. 5) which is the oldest greenhouse of the garden ( 6 ). The walkways have a width of 7.5 meters including two small sideways ground corridors with small plants, and are delimited by walls (fig. 3). In the first phase of our experimentation we are restricting the robot to move along the main walkways. The continuity of the walls is interrupted by either regularly spaced passages to the sectors of the garden or, occasionally, by trunks and protruding roots (4). The surface of the garden is composed of tuff powder and is pleasant enough for a human to ride. The same cannot be unfortunately be said about the small wheeled suspensionsless robot we introduce in the following section. Fig. 2. The Pioneer 3 AT robot at the Botanical Garden 2.2 The Robot The robotic platform we are using for the project is a Pioneer 3-AT from Mobile Robots (fig. 2) with several add-on sensors. The robot is a box-shaped (50 cm x 49 cm x 26 cm) compact unit with rounded front and rear sides. It weighs about 25 kg. Two motor sets, one for each side, powers four outdoor wheels with a diameter of 21 cm. Steering is accomplished by skid-driving motors in opposite ways; this makes the robot able to turn on place. Encoders on the motors axles and an inertial unit based on solid state gyroscope and accelerometers provide data for the odometry estimation. An electronic compass is also mounted on the robot. The onboard controller from Mobile Robots takes care of driving motors,

4 4 Antonio Chella, Irene Macaluso, Daniele Peri, and Lorenzo Riano performing all the odometry-related computations, and providing an high level interface to the robot. An EBX form-factor single board PC, powered by a Pentium Mobile processor clocked at 1.6GHz is also mounted on board. The PC is responsible for the higher level tasks, including localization and mapping, navigation, vision, sound output and speech. The PC has also the duty to deal with the laser rangefinder and GPS units. A Linux based distribution (Gentoo) is the operating system we chose to install on the onboard computer. Wireless ethernet can be used to connect to the robot. External amplified speakers deliver sound output loud enough to be audible in the open-air space. A Propak+ unit from Novatel provides GPS positioning. A fixed GPS base station located in a certified location nearby the garden can be paired with the onboard GPS unit by means of wireless modems to gain the robot the more precise Differential GPS positioning. The robot mounts frontally a Sick PLS Laser rangefinder, with 180 degrees aperture, that provides either 180 samples readings at 10 Hz, or 360 samples readings at 5 Hz. The measurements are in the range 0.5m-50m with 1cm accuracy. Ultrasound sonars on front and rear sides and Infrared proximity sensors provide further data for obstacle detection. A Bumblebee color stereo camera atop a pan-tilt unit gives robot the ability to take stereo pairs of color pictures (fig. 8). Three 12 Amps hour batteries gives the robot a maximum overall autonomy of about one hour and half. Fig. 3. RoBotanic riding on past the main crossing. 2.3 The experimental tour Our current goal is to make RoBotanic able to guide a simple tour in the selected area of the botanical garden (The System of Lynnaeus ) exposing visitors to Historical, Botanical and Architectural information by mean of its synthetic

5 RoBotanic: a Robot Guide for Botanical Gardens. Early steps. 5 voice. The tour starts in the half-circular area near the Gymnasium where visitors are introduced to the history of the garden, and the main buildings complex is described. The group then moves along the main walkway headed to the central crossing. During the walk informations concerning the sectors at the side of the walkway are given to the visitors. After reaching the crossing the group stops to be briefed about the intersecting walkway. The walk then resumes headed towards the Giardino d Inverno where the description of the historical greenhouse and the plants it contains is provided to the group of visitors (fig. 5). Then the walk continues back to the crossing where a right turn brings the group back on the main walkway headed to the Aquarium (fig. 3). Again along the walk the group is let to know about the remaining two sectors of the System until the round pool with aquatic plants is reached. Then, after the description of the Aquarium, the guide tells briefly visitors about the other parts of the garden. Finally the robot guides the group back to the starting point (fig. 4), and from there to the nearby exit gate where the visit ends. In this task RoBotanic is facing several specific challenges arising from the environment of the botanical garden as pointed out in the following section. Fig. 4. RoBotanic coming back from the Aquarium. Roots partially obstructing the walkway are visible on the right. 3 Early steps of a small robotic outdoor guide We have so far performed a large number of experiments collecting many detailed logs of the odometry, sonars and laser rangefinders data. We are thus able to make some early considerations. At first, we focused on building a map of the RoBotanic environment in order to test the robot in autonomous navigation accordingly with the process is described in [5] and [6]. For this early experiments

6 6 Antonio Chella, Irene Macaluso, Daniele Peri, and Lorenzo Riano Fig. 5. RoBotanic with the main greenhouse in the background. we used the Aria software by Mobile Robots in both the logging and mapping phases. We drove the robot along a continuous path covering the main walkways in their whole extent while recording a log of the sensor data (fig. 6). The result of the off-line mapping process [8] was correct in the walkways but shows sensible misalignment of the orthogonal segments and a short ghost walkway directed to north-west (fig. 7). The problem ariso mostly from systematic errors we are now able to account for and correct. Of course the tuff powder and irregularity of the surface of walkways are not quite helping in getting good odometry data. Other problems we realized -actually not quite unexpectedly- concern the Differential GPS that could boost considerably the mapping process if only was not hindered by the intricate leafage of tall trees and the stereo camera (fig. 8). Minor problems are generated by few protruding roots that partially obstruct the walkways (4) as they are generally detected by the rangefinder and included in the map(7). Our early experimentation continued with a map assisted navigation test that apart from some unexpected localization [7] failures gave us some more hints to improve our approach. 4 Conclusions We presented the early steps of our RoBotanic project describing the path we are following to deliver an outdoor robotic guide. While we think RoBotanic is the natural evolution of Cicerobot, most of the robotic architecture we have based our work so far needs some overhaul in the light of the experimental data we have collected so far. That is not unexpected considering the scarce literature on outdoor robotic guides and even the problems to adapt navigation, mapping, localization, and vision techniques that have been tested in indoor environments to the realm of outdoor robotics.

7 RoBotanic: a Robot Guide for Botanical Gardens. Early steps. 7 Fig. 6. Raw mapping of the entire walkways generated by driving the robot from the Giardino d Inverno for the entire walkway, then back halfway to the central crossing, then to the Gymnasium and finally along the entire main walkway down-to the Aquarium. Fig. 7. Map generated by the Mapper application. The map is correct in the walkways but shows sensible misalignment of the orthogonal segments, and shows a short ghost walkway directed to north-west.

8 8 Antonio Chella, Irene Macaluso, Daniele Peri, and Lorenzo Riano Fig. 8. Image taken by the on-board camera. Note the difficult lighting conditions and the abundant leafage that makes problematic receiving GPS signals. Acknowledgments. Authors would like to thank Prof. F. M. Raimondo, director of the Botanical Garden of the University of Palermo, and Dr. M. Speciale and Dr. M. Surano from the same institution, for their precious help and support. References 1. W. Burgard, A. B. Cremers, D. Fox, D. Hähnel, G. Lakemeyer, D. Schulz, W. Steiner, and S. Thrun. Experiences with an interactive museum tour-guide robot. Artificial Intelligence, 114:3 55, A. Chella, M. Frixione, and S. Gaglio. A cognitive architecture for artificial vision. Artificial Intelligence, 89:73 111, A. Chella, M. Frixione, and S. Gaglio. Understanding dynamic scenes. Artificial Intelligence, 123:89 132, A. Chella, M. Liotta, I. Macaluso, and Daniele Peri. Lessons learned with cicerobot, a robot for museum guided tours. In Eurobot 2006, J.A. Meyer and D. Filliat. Map-based navigation in mobile robots: I. a review of map learning and path-planning strategies. Elsevier Science, J.A. Meyer and D. Filliat. Map-based navigation in mobile robots: Ii. a review of map learning and path-planning strategies. Cognitive Systems Research, 4: , S. Thrun. Learning metric-topological for indoor mobile robot navigation. Artificial Intelligence, 99(1):21 71, B. Yamauchi and R. Beer. Spatial learning for navigation in dynamic environments. IEEE Transactions on Systems, Man and Cybernetics-Part B., 26(3): , Special Issue on Learning Autonomous Robots.