6. Conclusions and Future Research

This chapter briefly summarizes the major results of this study, discusses their significance and potential applications, and outlines suggestions for future work.

6.1 Conclusions

The work reported in this thesis enables us to summarize the following ideas:

• Modeling Army-ant robots as a self-organizing system has many advantages. Adaptive, collective and "complex" systems resulting from simple individual behavior are what the Army-ant scenario envisages. Simplicity of the individual agents is an important factor in implementation. However, the size and nonlinear character of self-organization leaves little possibility for definitive analysis.
• It is possible to geometrically arrange Army-ant robots by using a distributed approach. Robots can spatially organize themselves around a goal using only local information transferred by broadcast signals. This method is more advantageous (and faster) than a conventional centralized control methods, especially when the number of agents is large. The methods described in the text can be applied directly to other multirobot systems in underwater, planetary surface and space missions.
• It is also possible to separate the agents into different teams around different goals. The size of these teams can be determined by the difficulty of the assigned task. The team formation can again be achieved by using broadcast signals. Although there is no hierarchy between agents, temporary "leader" assignments seem to be necessary to overcome several problems. However, this not a violation of the homogeneous character of the population since all agents may became one and replace the leader.
• Use of communication channels enables cooperating robots to form a decision mechanism. Army-ant robots can share individual information using the coupled VDP oscillator scheme, and consequently "act" intelligent. The method described for collective load bearing may have several other applications in the Army-ant scenario.
• Driven by several behavioral modules, Army-ant robots form a large dynamic system. Interaction "rules" between agents have to be adjusted (or have to self-adjust) carefully to the environment and/or tasks to define the "responsibilities" of agents during different phases. This necessary for system robustness.
• The robot beacon signals have to carry some information other than indicating the relative position of the agent. Army-ant robots must signal their "status" to their teammates in order to find goals in areas with size larger than the detection region of the agents and in warehouses with "maze" structure. These signals are to be used in a similar way pheromone fields are used in insect societies.
• Self-organizing mobile robots need to be equipped with communication devices as well as beacons and detectors. Binary coded signals transferred at RF may prove to be highly useful at different stages of the Army-ant scenario. Implementation of such a system with broadcast characteristics is feasible. However, the computation of the distance to beacon(s) is a problem to be solved, considering that there will be many interfering signals.

6.2 Areas of Further Development

Several topics briefly mentioned in the text deserve further attention:

• Simulations described in Chapter 3 do not take into account parameters such as inertial constraints of the agents. The algorithms devised here can be extended to more complex programs for more realistic simulations incorporating robots' characteristics and collision avoidance. Such algorithms may be realized by object-oriented programming in an X-window environment.
• Using Lego Dacta robots to test the performance of the algorithms in the text may be very informative. Performance analysis under obstacle avoidance constraints can be analyzed better using actual mobile robots than computer simulations. Such an implementation would also provide useful information about signal interference and distance and/or direction detection.
• Topics related to multiagent systems, such as Petri nets, cellular automata, Kohonen networks and genetic algorithms, have to investigated for possible applications to Army-ant scenario. Petri net and cellular automata techniques can have applications in understanding the swarm behavior resulting for individual models. Again, genetic algorithm applications may be suitable for changing the behavior models (i.e. parameters acting on behavior modules and connecting them) to create an adaptive system.
• The use and feasibility of non-linear "heartbeats" in Army-ant robots have to be explored in depth. The variable size of robot teams may be a problem for oscillator couplings.
• The use of gas sensors as a distant alternative to radio and acoustic signals must be kept in mind. On going research on detection of specific odors may prove to be useful in mobile multirobot systems since pheromone fields play an important role in self-organizing insects.

Back to Thesis Main Page

To Chapter 5

To Appendix