Greg Detre
@17.30 Monday, 05 March, 2001
PhD at Stanford, moved to Xeroc PARC
simple modules, combine to achieve complex tasks, rearrange to form many configurations
versatile
robust (high redundancy)
inexpensive (economies of scale/batch production)
want a system that can rearrange itself
what are the effects/limitations? does increasing the number of modules increase what the robot can offer?
self-reconfigurable (mobile, lattice, closed chain) and modular
mobile � move around, then re-dock
lattice � move to different positions within a discrete lattice
closed chain � large, long chains of models, that change their shape
they�re good for versatility � where you don�t know what to expect
xerox � PR, component technologies and smart matter distributed control
funded by DARPA � interested in search & rescue (e.g. urban)
could do some of a difficult obstacle course
could:
turn/backwards, pipes, up/down stairs, pu slopes, loose debree, 0.5km on one battery chair
couldn�t:
ladder, vertical pipe, �
climbing like a caterpillar, climbing the stairs, riding a tricycle, spider (COMDEX 99)
very difficult inverse kinematics
dextrous workspace � moving around one fixed point vs 6D calculations
what if you had a really really long coiled robot and you wanted to convert it into a complicated structure � very difficult
telecube � similar to Crystalline (Dartmouth), but 3D � modules expand/shrink and attach/detach
very difficult to get the robot to move another object
considering genetic programming
proteo � smaller, rhombic dodecahedrons, more of them
but the actuation mechanism that could roll around the edges proved much harder than expected
instead, used magnets and humans to manipulate the modules like clay, and the workstation works out what shape has been formed
generation 3 (200 modules = �500,000)
self-diagnosis and self-repair
fully autonomous locomotoin in a rubble pile, going from 3 different shapes
need an integrated environment understanding
shrink modules further, increase numbers (cost becoming prohibitive)
explore automatic methods of control
learning based control
biological metaphors
specialised modules for some tasks
exploit general reconfigurability � do you need general reconfigurability, or might just a few shapes be enough on their own
considering bicycle-riding!
Summer interns
considered centralised module??? what are the adventages of distributed control???
every module will probably need some processing � having all the processing in just one place wouldn�t be efficient
they will employ a brain segment for path-finding etc.
single torque model � what about high torque vs high speed modes???
meta-module
what about robots that can hop???
they didn�t like that idea. they prefer low-speed high-torque systems
what about freedom in another dimension, e.g. twisting/turning???
all they can do is move in one degree of freedom � they have to be re-attached (which they can do themselves in the latest generations) to move in other than a snake pattern
what about manipulation?
snake form is best for locomotion, but not for manipulation. the robots won�t need to move human bodies
large + small-scale versions???
small-scale is more interesting for the moment, because you can more easily increase numbers
large ones could shore up and reinforce damaged buildings
submersible and space � don�t have to worry about torque and gravity???
yes, but there are always complications and technical issues
how tightly do you have to specify how the robot can recombine???s
can they reconnect on the fly???
what about tiny versions???
what about add-ons??? hands, pads etc.???