Moore s Law Claytronics Seth Copen Goldstein www.cs.cmu.edu/~claytronics RSS 6/11/05 Goodness Joint work with Todd Mowry and Baker, Campbell, Gibbons, Guestrin, Hoburg, Kufner, Lee, Pillai, Seshan, Sitti, Sukthankar, Veloso, Kirby, Aksak, Bhat, Bowers, DeRosa, Rister, Stanos Time RSS 6/05 2003-5 Goldstein&Mowry 1 RSS 6/05 2003-5 Goldstein&Mowry 2 Moore s Law? Where are we in 50 years? Goodness Time H Happy B B Day B day Day Cost Weight Volume Power Cycle time Storage 1949 Eniac 5M-23M (2002 $) 30 tons 450 M 3 200KW >200µs <800B 2003 greeting card 1$ 1 oz 1 cm 3 20mW 25ns 4KB 2050 Programmable matter 1 millicent 20 µg 1 nm 3?? (1 µm 3 ) 2 attowatts 2 picosec 16KB Cogent arguments for both sooner and later exist RSS 6/05 2003-5 Goldstein&Mowry 3 RSS 6/05 2003-5 Goldstein&Mowry 4
audio Encoding 01110010 mp3 pario Latin: to bear, bring forth, produce; create, make, get video Encoding 01110010 mpeg Encoding 01110010 RSS 6/05 2003-5 Goldstein&Mowry 5 RSS 6/05 2003-5 Goldstein&Mowry 6 Science Fiction? Fearing/UCB Sitti/CMU Science fiction? MEMS/Nanotech Amorphous Computing/ Emergent Behavior Sensor Nets Rus/MIT Modular Robots Materials Amorphous Computing/MIT Multi-Robot Teams klavins/uofw Yim/Parc Flynn/SRI Rabinett&BBC RSS 6/05 2003-5 Goldstein&Mowry 7 Flynn/SRI IRobot Veloso/CMU Claytronics/CMU RSS 6/05 2003-5 Goldstein&Mowry 8
Claytronics & Catoms Claytronics & Catoms The old way: Monolithic Not AI Just the system to render 3D dynamic objects Terminator 2: Judgment day trailer The Claytronic way: Ensemble of claytronic atoms The Claytronic way: Ensemble of claytronic atoms RSS 6/05 2003-5 Goldstein&Mowry 9 RSS 6/05 2003-5 Goldstein&Mowry 10 Ensemble of elements which each contain Processor Communication Power Sensing Display Actuation Claytronics & Catoms Programmable adhesive Can be programmed to form interesting dynamic shapes and configurations. A system for exploring the The computer Claytronic way: science of programmable Ensemble matter of claytronic atoms Capture 1. Capture 3D Object 2. Encode 3D model 3. Transmit data RSS 6/05 2003-5 Goldstein&Mowry 11 RSS 6/05 2003-5 Goldstein&Mowry 12
Reproduce A Claytronic Atom: Catom 3-5 years 5+ years RSS 6/05 2003-5 Goldstein&Mowry 13 RSS 6/05 2003-5 Goldstein&Mowry 14 A Claytronic Atom: Catom A Claytronic Atom: Catom The outside is studded with contacts Each catom can display an image/pixel RSS 6/05 2003-5 Goldstein&Mowry 15 RSS 6/05 2003-5 Goldstein&Mowry 16
Inside the Catom Inside the Catom RSS 6/05 2003-5 Goldstein&Mowry 17 RSS 6/05 2003-5 Goldstein&Mowry 18 Inside the Catom Moving the Catom Communication Antenna SuperCap Computer Programmable Magnets We can buy these today! RSS 6/05 2003-5 Goldstein&Mowry 19 RSS 6/05 2003-5 Goldstein&Mowry 20
Moving the Catom Moving the Catom RSS 6/05 2003-5 Goldstein&Mowry 21 RSS 6/05 2003-5 Goldstein&Mowry 22 Moving the Catom Simulation of Future Catoms RSS 6/05 2003-5 Goldstein&Mowry 23 RSS 6/05 2003-5 Goldstein&Mowry 24
Claytronics REV4 REV3 2D system Modular design RSS 6/05 2003-5 Goldstein&Mowry 25 RSS 6/05 2003-5 Goldstein&Mowry Magnets For Locomotion 26 Design Principles Goal: Scaling Up down (in numbers) (in size) sub-goals: RSS 6/05 2003-5 Goldstein&Mowry 27 RSS 6/05 simplicity robustness homogeneity no static power distributed, but catom local computation low complexity algs 2003-5 Goldstein&Mowry 28
ε Goal: Scaling Design Principles Up (in numbers) Some consequences: down (in size) No moving parts sub-goals: simplicity Genderless, unary connectors robustness No self-contained power source homogeneity New programming paradigm no static power distributed, but catom local computation low complexity algs Goal: Scaling The Ensemble Axiom Up (in numbers) down (in size) Consequences: No moving parts Genderless, unary connectors No self-contained power source New programming paradigm The Ensemble Axiom: A catom should include only enough functionality to contribute to the desired functionality of the ensemble RSS 6/05 2003-5 Goldstein&Mowry 29 RSS 6/05 2003-5 Goldstein&Mowry 30 A day in the life of a catom Get power! Localize Establish a network Get to work move to desired location If on surface, display video, The Robot is the tether Catoms have no battery They cooperatively create a power grid Challenge: connectors are unary genderless Requires two indep sub-graphs + or - unary insulator + binary V dd Ground RSS 6/05 2003-5 Goldstein&Mowry 31 RSS 6/05 2003-5 Goldstein&Mowry 32
The Robot is the tether Catoms have no battery They cooperatively create a power grid Challenge: connectors are unary genderless Requires two indep sub-graphs Multiphase Algorithm First, passive resister net Second, distributed environment/guid-oblivious algorithm Third, fix-up 20v portion of a robot ensemble V dd Ground RSS 6/05 2003-5 Goldstein&Mowry 33 RSS 6/05 2003-5 Goldstein&Mowry 34 Multiphase Algorithm First, passive resister net Second, distributed environment/guid-oblivious algorithm Third, fix-up Multiphase Algorithm First, passive resister net Second, distributed environment/guid-oblivious algorithm Third, fix-up RSS 6/05 2003-5 Goldstein&Mowry 35 RSS 6/05 2003-5 Goldstein&Mowry 36
Algorithms Scale Distributed Localization Scales sublinearly with diagonal of ensemble RSS 6/05 2003-5 Goldstein&Mowry 37 RSS 6/05 2003-5 Goldstein&Mowry 38 Next Step, Create Network Use simple local rules to form hierarchy 10 line program does this! Local only decisions Global effect Getting There From Here Goal: Robust ensemble of millions of catoms Claytronics Design Principles No Moving Parts Local Control No Static Power time RSS 6/05 2003-5 Goldstein&Mowry 39 RSS 6/05 2003-5 Goldstein&Mowry 40
Proposal for 3D asic Dimensions Weight power Locomotive mechanism Scaling of Claytronics Macro Micro Nano >1 cm >1 mm <10 microns 10 s gr 100 s mg <1 mg <2 Watts 10 s mw 10 s nw Programmable Electrostatics Aerosol magnets FEM simulation: Flux 150 A-Turns 10mm Catom Catom diameter 44mm 20mm 10mm 10mm Amp- Turns 260 150 250 150 Power (W) 450 RSS 6/05 2003-5 Goldstein&Mowry 41 10 3.3 6.6 2.8 Force (mn) 78 50 33 Weight (mn) 6420 192 24 24 Electromagnets Adhesion Nanofiber Programmable Molecular surface mechanism adhesives nanofiber adhesion and adhesives covalent bonds Magnets Manufacturing methods Conventional manufacturing and Micro/Nanofabrication and Chemically directed self- assembly micro-assembly assembly and fabrication Resolution Low High High Cost $$$/catom $/catom Millicents/catom RSS 6/05 2003-5 Goldstein&Mowry 42 What about the software? Programming Models Distributed Planning Networking OS One example: Motion planning Motion Planning Challenges Large Number of Catoms ( ~ 100k+) No central planner Physical Limitations Maximum torque Center of gravity Structural stability Power/network connectivity Inaccuracies/failures RSS 6/05 2003-5 Goldstein&Mowry 43 RSS 6/05 2003-5 Goldstein&Mowry 44
Hole Flow Methods Based on concept from moving tile puzzles or semiconductor physics Planning is scale-independent Three primitives move create destroy Motion Planning with Holes Create raises contour RSS 6/05 2003-5 Goldstein&Mowry 45 RSS 6/05 2003-5 Goldstein&Mowry 46 Motion Planning with Holes Create raises contour Destroy lowers contour Holes circulate freely through structure (Brownian motion/gas molecules) Motion Planning with Holes Create raises contour Destroy lowers contour Holes circulate freely through structure (Brownian motion/gas molecules) Grow Grow shrink RSS 6/05 2003-5 Goldstein&Mowry 47 shrink RSS 6/05 2003-5 Goldstein&Mowry 48
Keeping it local and smooth Effect of criticality smoothing Hole motion based on local rules How do we control the global effect? Use ideas from self-organized criticality to do local Collapsing/smoothing RSS 6/05 2003-5 Goldstein&Mowry 49 RSS 6/05 2003-5 Goldstein&Mowry 50 Making a shape Claytronics & Pario Open up an entire new application space Antennas (Programmable Antennas) Design (100x protein model) Entertainment (interactive clay) Interaction (telepario) Rescue (paramedic on demand) Metal Man (fault tolerant robotics) Vehicle for studying CS problem of the future: How do you design, program, maintain, and use a billion component system? RSS 6/05 2003-5 Goldstein&Mowry 51 RSS 6/05 2003-5 Goldstein&Mowry 52
Claytronics Open up an entire new application space Vehicle for studying CS problem of the future: How do you design, program, maintain, and use a billion component system? Our Approach: The Ensemble Axiom Make scaling work for us Exploit scale invariance Design for scalability in both number & size: Millions of Micron-scale catoms RSS 6/05 2003-5 Goldstein&Mowry 53