256 CATALYZING INQUIRY
highly distributed sensor system.^22 Each dust mote has sensors, processors, and wireless communica-
tions capabilities and is light enough to be carried by air currents. Sensors could monitor the immediate
environment for light, sound, temperature, magnetic or electric fields, acceleration, pressure, humidity,
selected chemicals, and other kinds of information, and the motes, when interrogated, would send the
data over kilometer-scale ranges to a central base station, as well as communicate with local neighbors.
This architecture was the basis of an experiment that sought to track vehicles with an unmanned aerial
vehicle (UAV)-delivered sensor network.^23 The prototype sensors were approximately a cubic inch in
volume and contained magnetic sensors for detecting vehicles (at ranges of about 10 meters), a micropro-
cessor, radio-frequency communications, and a battery or solar cell for power. With six to eight air-
delivered sensor motes landed diagonally across a road at about 5-meter intervals, the sensor network was
able to detect and track vehicles passing through the network, store the information, and then transfer
vehicle track information from the ground network to the interrogating UAV and then to the base camp.
The subsumption architecture also asserts that this robust behavior can emerge from the bottom
up.^24 For example, in considering the problem of an autonomously functioning vehicle (i.e., one that
drives itself), a series of layers can be defined that
- Avoid contact with objects (whether the objects move or are stationary),
- Wander aimlessly around without hitting things, and
- Explore the world by seeing places in the distance that look reachable and heading for them.
Any given level contains as a subset (subsumes) the lower levels of competence, and each level can
be built as a completely separate component and added to existing layers to achieve higher levels of
competence. In particular, a level 0 machine would be built that simply avoided contact with objects. A
level 1 machine could be built by adding another control layer that monitors data paths in the level 0
layer and inserts data onto the level 0 data paths, thereby subsuming the normal data flow of level 0.
More complex behavior is thus built on top of simpler behaviors.
Brooks claims that the subsumption architecture is capable of accounting for the behavior of insects,
such as a house fly, using a combination of simple machines with no central control, no shared representa-
tion, slow switching rates, and low-bandwidth communication. This results in robust and reliable behavior
despite its limited sensing capability and an unpredictable environment, because individual behaviors can
compensate for each others’ failures, resulting in coherent and emergent behavior despite the limitations of
the component behaviors. A number of robots have been built using subsumption architectures. Of particu-
lar note is Hannibal,^25 a hexapod with more than 100 physical sensors and 1,500 augmented finite-state
machines grouped into several dozen behaviors split over eight on-board computers.^26
8.2.3 Robotics 2: Bacterium-inspired Chemotaxis in Robots^27
The problem of locating gradient sources and tracking them over time is an important problem in
many real-world contexts. For example, fires cause temperature gradients in their immediate vicinity;
(^22) See, for example, http://robotics.eecs.berkeley.edu/~pister/SmartDust/.
(^23) See http://robotics.eecs.berkeley.edu/~pister/29Palms0103/.
(^24) R.A. Brooks and A.M. Flynn, “Fast, Cheap and Out of Control,” 1989.
(^25) C. Ferrell, “Robust Agent Control of an Autonomous Robot with Many Sensors and Actuators,” Ph.D. thesis, Department of
Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, 1993.
(^26) A finite-state machine is a machine with a finite number of internal states that transitions from one state to another on the
basis of a specified function. That is, the argument of the function is the machine’s previous state, and the function’s output is its
new state. An augmented finite-state machine is a finite-state machine augmented with a timer that forces a transition after a
certain time.
(^27) Material in Section 8.2.3 is based on excerpts from A. Dhariwal, G.S. Sukhatme, and A.A.G. Requicha, “Bacterium-inspired
Robots for Environmental Monitoring,” International Conference on Robotics and Automation, New Orleans, LA, April 2004.