Slashdot Mirror
Linguistics Could Help Future Driverless Cars Cooperate Better (thestack.com)
An anonymous reader writes: A team of swarm robotics researchers have applied a linguistics technique typically used in manufacturing to automatically program and control a 600-strong robot fleet. The scientists found that human error was significantly reduced, making the solution safer and more reliable than previous 'trial and error' approaches. The tasks in the experiments were defined by a graphical tool, which a machine then automatically translated to the bots. The supervisory technique uses a linguistics system through which the robots construct their own 'words', related to what they can 'see' and which moves they choose to action next. Robots will only perform actions from valid 'words', which means they are guaranteed to carry out the required tasks.
Not me. Hopefully once the tech takes off, manual driving will be illegal for folks like you who don't want driverless cars. That will allow the rest of us a safe and efficient mode of transportation.
All my liberal friends think I'm a conservative, all my conservative friends think I'm a liberal.
First off, direct link to the corresponding journal article (open access).
The "linguistics technique" is apparently supervisory control theory. I'm not too familiar with it, but apparently in supervisory control theory you model both the capabilities of the robots and the goals you want as formal system, in the form of discrete states for each actor, and events which cause transitions between the states (i.e. the robot is a finite state machine). These events can either be controlled (the robot performs an action) or uncontrolled (something happens in the environment outside the robot's control).
The twist that supervisory control theory apparently brings to the party (and I'll admit I'm a bit unclear on this part) is that it encodes the FSM as generators which can propose possible sequences of events, and then looks for paths through the FSM which will bring you to the desired state. Paths which don't arrive at your desired state or are inconsistent with the structure of the FSM are trimmed. The linguistics part of it is that the paths are viewed as "words" composed of event "letters", and various linguistics theories are used to limit the space of possible "words" that are consistent with the "grammar" (the FSM structure) and which will arrive at your desired end conditions.
From what I understand, most of the processing is done up front, and the possible end states and paths to get there are machine translated to an actual control program that not only acts like it obeys the FSM, but also is able to direct it's actions toward the desired end state.
The paper is a little technically dense, though (and quite slanted toward formal method worship), so I'd appreciate any corrections/additions anyone with more experience would have.