Werger’s paper discusses that a strict behavior-based approach can scale to higher levels of complexity than many robotics researchers assume and that the resulting systems are in many cases more efficient and robust than those that rely on “classical AI” deliberative approaches. The focus of the paper is on the systems of cooperative autonomous robots in dynamic environments. The main point of the paper is to drive the point that the key to success in these areas lies in the philosophy of fully exploiting minimal systems. The benefits of minimalism, including easier and faster development and debugging and a more efficient and robust execution system, have led to recent popularity of minimalist systems.
To have a purely reactive architecture, there can not be any notion of saving state within this architecture. The interesting technique that they use is that since behaviors are all purely reactive, some of the inputs fade over time, rather than disappearing suddenly. The cool effect of this behavior is that when the robots were playing soccer, once the vision sensor started to fade and the robot lost the focus of the ball, the robot would turn away from the ball with its back to hit the ball. This produced some very interesting emergent behavior. The team did demonstrate the feasibility of a purely reactive team playing in the confines of a soccer game.
The maintenance of the design at the abstract behavior level was not done until final coding. They spent at least ninety percent of development time without any written code and were then able to code and debug system reliably in a matter of hours. The system demonstrated effective obstacle avoidance in a dynamic environment, generation of smooth, effective trajectories, three separate methods of ball handling and dynamic configuration into appropriate population-limited offensive and defensive formations. This architecture utilized strict behavior-based techniques with no explicit communication. This aspect of the architecture is fascinating because it truly models the behavior of hive minded insects in its interaction within the soccer tournament.
The team remained undefeated throughout the competition. In fact, no opposing team was able score a goal on Werger’s team through the entirety of the competition. Interestingly enough, one of the rules in the competition – avoid hitting other players – was thrown out. If the rule had remained in effect, then Werger’s team would have won.
In conclusion, it is interesting to see behavior based robots working collaboratively within the environment of this multi-agent system for a soccer match. I agree that it is important to research the robot interactions using the minimalist approach to truly understand how robots can interact at the most basic of levels. Understanding this layer will lead to more complex behavior later on by stacking additional behavior modules on top of the main one. By any objective measure, the success of Werger’s team was how they were able to demonstrate actual complex team behavior without explicit communication or off board processing and develop a robust obstacle avoidance algorithm using a minimalist behavior-based programming approach with limited resources and months of effort.
Reference:
- Werger, Barry Brian. “Cooperation without Deliberation: A Minimal Behavior-based Approach to Multi-robot Teams.” <http://www8.cs.umu.se/kurser/TDBD17/VT06/utdelat/Assignment%20Papers/Werger.pdf>
