Distributed Autonomous Robotic Systems 5. Asama H. et al.(eds), pp. 91-100. Springer-Verlag, Tokyo. (2002)
Autonomous Robots Sharing a Charging Station with no Communication: a Case Study Francois Semp´e1 , Angelica Mu˜ noz2 , and Alexis Drogoul2 1
2
France Telecom R et D, 38 du Gal Leclerc, 92131 Issy-les-Moulineaux, France.
[email protected] Laboratoire d’Informatique de Paris 6, 4 place Jussieu, 75252 Paris Cedex 05, France. angelica.munoz -
[email protected]
Abstract. This research focuses on the design of a group of self-sufficient mobile robots, such that a group of three robots can remain in operation and efficiently share a charging station, using simple mechanisms. An experimental bottom-up approach has been adopted in order to test various strategies to manage collective self-sufficiency, which rely upon low-level mechanisms such as non-direct communication and non-complex decision making. keywords: collective robotics, mobile robot, self-sufficiency, charging station, sharing strategy.
1
Introduction
This research focuses on the design of a group of self-sufficient mobile robots. Self-sufficiency denotes the ability of a system to maintain itself in a viable state for long periods of time [1], which means that a self-sufficient robot has to be able to ensure its power supply by itself. For that, it has at its disposal certain recharging facilities, i.e. rechargeable batteries and a selfrecharge device, and relies on several mechanisms so as to be able to examine its power supply constantly and to locate and use a charging station. The design of a group of self-sufficient robots introduces new challenges, because a robot that is part of a group not only has to ensure its own operation, but also has to cope with its partners and share common, usually essential resources with them. Self-sufficiency is at the core of the design of autonomous robots but surprisingly this topic has not been much addressed in the literature. Maybe it is not considered to be a noble issue but merely a technical one. Or maybe the specialized skills required to study the question of recharging batteries are not commonly found in robotic research teams. There is even less work on the self-sufficiency of a group of robots.
The research described in this paper is part of the MICRobES1 project, which aims at implementing a group of mobile robots which will be able to operate, or survive, permanently in the corridors of our laboratory. The paper describes the research to enable a group of three MICRobES robots to remain in operation and efficiently share a charging station, using simple mechanisms. An experimental bottom-up approach has been adopted in order to test various startegies to manage collective self-sufficiency which rely upon low-level mechanisms such as non-direct communication and non-complex decision making. The paper is organized as follows: section 2 adresses the problem of sharing a station by a group of robots and considers related work. Section 3 describes our proposal to design self-sufficient robots, and gives details of the hardware and the control of our robots, as well as the framework of the experiments. Section 4 discusses the results and section 5 presents conclusions and future lines of research.
2
Sharing a charging station
2.1
The problem: to survive and to be useful
What is it expected of a group of robots sharing a charging station? There are two main goals. First, all robots must survive for long periods of time; second, they need to share the station efficiently, in order to carry out certain tasks. What does efficient sharing mean? Let us reuse the concept of basic cycles presented by McFarland and Spier [2]. A basic cycle corresponds to the three stages a self-sufficient and useful robot has to repeat: to work, to reach and connect to the station and to recharge the battery. Roughly speaking, an efficient cycle maximizes the time spent working. How can that be done? Charging time is irrelevant here, as it cannot be modified by robot behaviors and depends mostly on charger speed. Supposing that it is constant, increasing working time means decreasing the time spent in reaching the station. For a single robot the best strategy will probably be to reduce the number of recharges by only trying to reach the station when the battery is low. For a group of robots, however another stage has to be considered: once a robot has reached the recharge area, it may have to wait because a partner has already occupied the station. Waiting time is a problem for two reasons: first, it represents a danger as it may lead to a flat battery and second, it is a waste of time that must be reduced. What does a group of robots sharing a charging station need? In order to make sharing possible, the charging station must be sufficiently fast with respect to the number of robots. Recharge speed refers to the autonomy 1
MICRobES is an acronym in French for Implementation of Robot Collectivities in a Social Environment.
provided by, let us say, one minute of recharge. For example, a recharge speed of 5 means that one minute of recharge will give a robot 5 minutes of autonomy. Such a charging station can theoretically supply 6 robots with the same autonomy: once the first robot has gained five minutes of autonomy, five one-minute slots are available for five more robots. But in practice, a charging robot cannot change places instantaneously with another one and the station must be free for a while to let the robots switch. In fact, the longer this free time, the easier the sharing, and it should not be very difficult for three robots to share a station with a recharge speed of fifty. On the other hand it is difficult, not to say impossible, to build such a station. What decisions do robots have to make? For the station-sharing problem, a robot has to take two major decisions: when to go and recharge and when to leave the station. There are many possible strategies, based on individual state, partner state, priority rules, negociation and so on. 2.2
Related work
The station-sharing problem has rarely been addressed in research and, as far as we know, no paper has been written just on this subject. Two research projects do exist, however. Steels [3] built an ecosystem where rivals have to compete for power supply. Two mobile robots share a station but no detailed results are presented. Michaud [4] suggested using artificial emotions in order to organize long term activity for a group of robots. However, no experiments involving physical robots have been presented yet. The questions of recharge and self-sufficiency for one robot are a bit more common. Birk [5] points out the problem of batteries and shows that cell chemistry may constrain robot behavior. Last, a kind of sport record has been established by Yuta ans Hada [6]. They made a robot that ran continuously for a week recharging its battery every ten minutes.
3
The proposal
This paper proposes a bottom-up experimental approach to the stationsharing problem. It has been shown that a group of robots exhibiting simple behaviors can achieve non-trivial tasks without the help of communication[?], and we want to explore this further and to identify the conditions for such a sharing method. 3.1
A basic strategy for sharing
What could be a basic strategy for a robot team to share a station? As already mentioned above, two decisions have to be taken by each robot: when to go and recharge and when to leave the station. One of the first ideas that comes
to mind is to set a go-and-recharge threshold that triggers the equivalent behavior. This threshold refers to an energy level. For instance, a 4-minute go-and-recharge threshold means that a robot tries to reach the station when less than four minutes of autonomy are left. But when should a robot leave the station? When it has reached a given level of energy. Thus all robots will leave the station with the same remaining autonomy. This is a basic strategy but why should it work? The assumption is that robots will alternate at the station because their power supply will reach the go-and-recharge threshold at different times. In our case, since there is only one station to share and each robot leaves the station with the same amount of power supply, their autonomy at a given time is always different. However, there is a bootstrap problem. What if the robots start with the same autonomy? They are going to rush to the station at the same time, which is probably a bad idea. We carried out two series of experiments on alternation: the first with robots starting with different energy levels, the second with robots starting with the same energy level. 3.2
Hardware
The experiments have been caried out using three Pioneer 2-DX mobile c provided with odometers, bumpers, sonars, rarobots from ActivMedia , dio modems and video cameras. We also have a charging station made by the LIP6 laboratory and France Telecom R & D. The robots have been modified in order to use this station. Both an electronic card to control the temperature and the current, and charging plates at the rear of the robots have been adapted. The original batteries have been replaced by lead batteries that accept high currents up to 60 Amps. The charging station has been designed to accept an approximate connection between it and the robots2 . Robots take from 20 to 30 seconds to reach and connect to the charging station that they have just located and the time needed to recharge depends on the current supply of a robot and the age of its battery; on average 15 minutes are sufficient to provide 2 hours of autonomy. The current supplied by the charger is very irregular. In order for experiments to be reproduced, the recharge speed percieved by the robot is constant, and below the actual but unpredictible one. 3.3
Control
Robots have a repertoire of basic behaviors that are activated by the external stimuli they perceive. Basic behaviors such as avoiding obstacles, wandering, navigation and localization have been designed for our research team. These behaviors have been successfully implemented in and tested on our robots, 2
Patent pending
but discussion is beyond the scope of this paper (some details can be found in [8]). In order to be self-sufficient, a robot has to examine its power supply constantly. When it perceives that it is below the go-and-recharge threshold, it goes towards the area where it knows that there is a charging station. If the robot recognizes that it is within this area, it starts to search for the exact position of the charging station by revolving around itself and goes in that direction when it perceives the visual landmark that identifies it. Then the robot adjusts its position to connect to the charging station, stays there while its battery is being recharged, and finally leaves the station to restart its loop. If a robot finds the station occupied, it wanders for 15 seconds and tries to connect again. Figure 1 illustrates the behavior described, for a robot whose main task is to wander. start
Wander (15 sec min)
station left
battery
low
GoToRe− chargeArea
LeaveChar− gingStation
area found station not seen battery recharged
Charge
station not reached
station reached
ReachChar− gingStation
SearchForCh− argingStation station seen
Fig. 1. Automaton that summarizes the behavior of an individual robot.
3.4
Experimental settings
The environment, about 20 square meter in area, is L-shaped i.e. since the station is not always in sight, navigation is necessary. Three Pioneers start at the same time. Each run lasts one hour at most or until a robot dies. The maximum autonomy has been set to 15 minutes in order to have more cycles. Two go-and-recharge thresholds are tested: 4 and 7.5 minutes. These values are discussed later. The recharge speed represents a critical parameter, as underlined earlier. In our experiments, the charger gives approximately 5 minutes of autonomy for 1 minute of recharge. As three robots work together, the station will be free more or less half of the time.
4
Results and discussion
4.1
Maintaining alternation
In all the experiments presented in this section, the three robots start with 5, 10 and 15 minutes of autonomy respectively. Thus, at the beginning alternation at the station should occur; the questions being the maintenance of alternation and its efficiency.
Autonomy (minute)
The basic strategy. Two experiments were done with a go-and-recharge threshold set at 7.5 minutes. In both cases, robots managed to survive for one hour before we interrupted the experiments. They alternated at the station but irregularly. Figure 2 shows the energy level of the three robots during run 1. We can see that the robots started their charging stages (increasing functions) at very different levels, from 7 to 2 i.e. there were often long delays between the moment they triggered the go-and-recharge behavior and the moment they connected themselves to the station. Delays have one major cause: interference. This is because a working robot may wander into the recharge area and may stand for a while between the station and another robot that is trying to reach the station or connect itself. Consequently this robot will fail, either because station landmark is hidden or because it cannot get access to the station.
Robot 1 2 3
15
10
5
0 0
500
1000
1500 2000 Time step (one-hour run)
2500
3000
3500
Fig. 2. The autonomy of the 3 robots with a 7.5-minute go-and-recharge threshold.
Sanctuary strategy. The interference problem is not surprising and it has already been pointed out [9]. In order to reduce failures caused by this phenomenon, we have chosen to prevent working robots from entering the station area. The benefit is clearly demonstrated if we compare the number of
recharge tries shown in Table 1. Using the sanctuary strategy, the number of failures to reach the station has dropped drastically, though not yet entirely. There are three reasons for this persistence. First a robot may try to reach a busy station. Second, interference still occurs with a robot leaving the station after its recharge stage. Third, identification of the station using the camera fails from time to time. The following strategies are based on sanctuary: working robots avoid the station area. Table 1. Total numbers of attempts and success in connecting to the station. Basic and sanctuary strategies are compared. strategy Basic Sanctuary
run1 run2 run1 run2
Total tries Successes 28 15 30 15 18 15 21 17
Sharp strategy. This strategy is used to improve the efficiency of sanctuary strategy by decreasing the go-and-recharge threshold. Two experiments have been caried out with a 4-minute threshold. The first important point is the success of these new runs; 4 minutes are enough to reach the station and connect to it despite remaining connection failures. Second, when sanctuary and sharp strategies are compared, we see that robots spend noticeably more time working if a sharp strategy is adopted, with a 5 pecentage point increase from 66.6% to 71.6%, on average. The explanation is simple: robots recharge themselves more rarely - approximately one time fewer in a one-hour run - and in consequence spend less time in reaching and connecting to the station. Another benefit must be mentioned: the average duration of working periods increased too. 4.2
Creating alternation
The failure of the sanctuary strategy. It has been shown above that the robots alternated efficiently at the station, thanks to the different initial energy levels. But what would happen if robots started with the same autonomy? In other words, how can the alternation be created? Let us now consider the case of robots that all start with 10 minutes of autonomy. No experiment is necessary to show that a 4-minute recharge threshold will lead to a disaster as more than two minutes are necessary to recharge one robot. The 7.5-minute threshold has been tested twice. In both cases one of the robots died before its first recharge: three robots cannot
recharge themselves within a period of 7.5 minutes. A negative side effect makes the situation worse: as the robots reached the threshold approximately at the same time, they all rushed together to the station, thus creating a lot of interferences. Increasing the threshold is not the right solution, since the robots has to spend most of their time working.
Opportunism : a way to create alternation. The new strategy has to exhibit two contradictory properties: greed and priority for a starving robot. On the one hand, when they start robots must try to go and recharge as soon as possible in order to use all their remaining autonomy for a vital purpose. On the other hand, a starving robot must access the station more easily than the others. But remember that no robot knows anything about the other robots’ state. Opportunism is the behavior that makes a robot recharge itself when it sees the station, whatever its remaining autonomy. In theory, associating opportunism with a go-and-recharge threshold regroups the two properties. First, greed, because robots may use the station before the go-and-recharge threshold is reached, i.e. before they really need energy. Secondly, starving robots have a better chance of reaching the station because their behavior is deterministic: even if they cannot see the station they stay around and check for it regularly. Figure 3 shows the power supply of an opportunistic strategy experiment. Opportunistic behavior may be triggered only when remaining autonomy is less than 10 minutes: we do not want too greedy robots. Alternation was successfully created and maintained. On this run, no robot failed to connect to the station: nineteen connection attempts led to nineteen recharges. None had to wait, since a robot charging its battery hides the station landmark, thus inhibiting opportunistic behavior. On the other hand, robots went and recharged very early, reducing the duration of working period (7.2 minutes on average). Opportunism is necessary only at the first stage, to create the alternation, after which the sharp strategy proved to be efficient. In order to take advantage of both strategies, a temporary opportunism was implemented. Robots are opportunistic for two cycles, and then adopt a pure sharp strategy. Two successful experiments have shown the creation of alternation and long working periods (more than 10 minutes). Last, figure 4 illustrates how opportunism can cope with robots starting with different energy levels. We can see at the beginning that the most starving robot has been overtaken by another one because of an opportunistic behavior. However the first one waited until the end of the recharge and connected the station in time.
Autonomy (minute)
Robot 1 2 3
15
10
5
0 0
500
1000
1500 2000 Time step (one-hour run)
2500
3000
3500
Autonomy (minute)
Fig. 3. The autonomy of the 3 robots with an opportunistic strategy.
Robot 1 2 3
15
10
5
0 0
500
1000
1500 2000 Time step (one-hour run)
2500
3000
3500
Fig. 4. The autonomy of the 3 robots with a temporary opportunistic strategy.
4.3
Discussion about the go-and-recharge threshold
The go-and-recharge threshold is at the core of the strategies presented above. This section discusses its properties in relation to two other parameters: the time necessary to reach the station and the average recharge time of one robot. Obviously, the go-and-recharge threshold must be set above the maximum duration of the navigation and connection phases. In an extensive environment, it should depend on the distance between the location of the robot and that of the charging station. Interference with other robots, navigation and connection problems must be taken into account; with our experimental settings a 4-minute threshold was enough. In our case, the navigation stage rarely exceeded 1 minute and the connection stage a mere 30 seconds. With a 4-minute threshold, this left more than 2 minutes of autonomy, which is the approximate duration of a recharge. Thus a robot can wait for another one to finish recharging, which is why the opportunistic strategy works.
5
Conclusions and future lines of research
This paper presented various strategies to enable a group of robots to share a charging station. These strategies are based on simple mechanisms that do not suppose communication between robots. The first ones use a go-and-recharge threshold: a robot tries to reach the station when its autonomy drops below a given value. The strategies are oriented to maintain the access to the station alternately and were tested on robots with non-equivalent energy needs. The last strategy enables robots with equivalent energy needs to balance their needs and access to the station alternately, through opportunistic recharge behavior. Results of the different strategies are presented and compared. Future work will focus on experiments with groups of more robots in more complex situations, in order to analyze the influence of the size of the group on the strategies, and to improve mechanisms to share a charging station.
References 1. McFarland D. (1995) Autonomy and Self-Sufficiency in Robots. The Artificial Life Route To Artificial Intelligence. Building Embodied, Situated Agents. Steels L.(ed). Lawrence Erlbaum Ass. Pub. USA, 187-213. 2. McFarland D., E. Spier. (1997) Basic Cycles, Utility and Opportunism in Selfsufficient Robots. Robotics and Autonomous System (20), 179-190. 3. Steels L. (1994) A case study in the behavior-oriented design of autonomous agents. In: From animals to animats 3. Proceedings of the 3rd International Conference on Simulation of Adaptive Behavior. Clif D., Husbands P., Meyer J.-A., Wilson S.W. (eds). USA, MIT Press, 445-452. 4. Michaud F., Robichaud E., Audet J. (2001) Using Motives and Artificial Emotions for Prolonged Activity of a Group of Autonomous Robots. To appear in: Proceedings of the AAAI Fall Symposium on Emotions. Cape Code Massachussetts, USA. 5. Birk A. (1997) Autonomous Recharging of Mobile Robots. In: Proceedings of the 30th International Sysposium on Automative Technology and Automation. Isata Press. 6. Yuta S., Hada Y. (2000) First Stage Experiments of Long Term Activity of Autonomous Mobile Robot: Result of Repetitive Base Docking over a Week. In: Proceedings of ISER’00, 7th International Symposium on Experimental Robotics, 235-244. 7. Beckers R., Holland O.E., Deneubourg J.-L. (1994) From local actions to global taks: Stigmergy and collective robotics. In Artificial Life IV, Brooks R., Maes P. (eds), USA, MIT Press, 181-189. 8. Viel P.-E., Drogoul A., Milgram M. (2001) Localization and Identificatoin: a Robust Landmark Recognition Systems for Mobile Robots. Technical Report LIP6, UPMC, France. 9. Goldberg D., Mataric M. Interference as a Tool for Designing and Evaluating Multi-Robot Controllers. In: Proceedings of the AAAI-97 Conference. Providence, Rhode Island (1997) 637-642.
