Cog's Computational System
The Cog Shop
MIT Artificial Intelligence Laboratory
545 Technology Square, #920
Cambridge, MA 02139
The computational control for Cog is
a heterogeneous network of many different processors types operating at different levels
in the control hierarchy, ranging from small microcontrollers for joint-level control to
digital signal processor(DSP) networks for audio and visual preprocessing.Cog's brain has
undergone a series of revisions.
The original was a network of 16 MHz Motorola 68332 microcontrollers on
custom-built boards, connected through dual-port RAM. Each of these nodes ran L, a
multithreading subset of Common Lisp.
The current core is a network of 200 MHz industrial PC computers running
the QNX real-time operating system and connected by 100VG ethernet. The network currently
contains 4 nodes, but can be expanded at will by plugging new nodes into the network hub.
QNX provides transparent and fault-tolerant interprocess communication over the network.
The PC backplanes provide ample room for installing commercial or custom I/O boards and
controller cards. The old and new brains can inter-operate, communicating via custom-built
shared memory ISA interface card.Video and audio preprocessing is performed by a separate
network of Texas Instruments C40 digital signal processors which communicate via the
proprietary C40 communications port interface. The network includes C40-based
framegrabbers, display boards, and audio I/O ports. The processors relay data to the core
processor network via ISA and PCI interface cards.Each joint on the robot has a dedicated
local motor controller, a custom-built board with a Motorola HC11 microcontroller, which
processes encoder and analog inputs, performs servo calculations, and drives the motor via
pulse-width modulation. For the arms, the microcontroller generates a virtual spring
behavior at 1kHz, based on torque feedback from strain gauges in the joints.
Brought to you by, Matthew Marjanovic
The primary force behind Cog's new computational system.
Representatives of the press who are interested in acquiring further
information about the Cog project should contact Elizabeth Thomson, email@example.com, from the MIT News Office, http://web.mit.edu/newsoffice/www/ .