1. Printing Functional Systems
Worlds Within Worlds
Hod Lipson
Mechanical & Aerospace Engineering
Computing & Information Science
Cornell University
Computational Synthesis Lab
Cornell University
College of Engineering

2. Adaptation Changing environments, tasks, internal structures
Behavioral adaptation
Morphological adaptation

3. Breeding machines in simulation
Lipson & Pollack, Nature 406, 2000

5. Emergent Self-Model
Bongrad, Zykov, Lipson (2006) Science, in press

6. Damage Recovery
With Josh Bongard and Victor Zykov

7. Making Morphological Changes in Reality

8. Printable Machines

9. Multi-material processes
Continuous paths
Volume Fill
High-resolution patterning, mixing
Thin films (60nm)

10. Illustration: Bryan Christie
Multi-material RP
Illustration: Bryan Christie

11. Our RP Platform
Fabrication platform: (a) Gantry robot for deposition, and articulated robot for tool changing, (b) continues wire-feed tool (ABS, alloys), (c) Cartridge/syringe tool

12. Printed Active Materials
Some of our printed electromechanical / biological components: (a) elastic joint (b) zinc-air battery (c) metal-alloy wires, (d) IPMC actuator, (e) polymer field-effect transistor, (f) thermoplastic and elastomer parts, (g) cartilage cell-seeded implant in shape of sheep meniscus from CT scan.
With Evan Malone

13. Zinc-Air Batteries
With Megan Berry

14. Zinc-Air Batteries

15. IPMC Actuators

16. IPMC: Ionomer Ionomeric Polymer-Metal Composite “Ionic polymer”
Branched PTFE polymer
Anion-terminated branches.
Small cation

17. First printed dry actuator
Quantitative characterization
Improve service life
Reduce solvent loss
Reduce internal shorting
Improve force output, actuation speed

18. Embedded Strain Gages
Silver-doped silicon
Robot finger sensor

19. IPMC: Ionomer Ionomeric Polymer-Metal Composite “Ionic polymer”
Branched PTFE polymer
Anion-terminated branches.
Small cation

20. First printed dry actuator
Quantitative characterization
Improve service life
Reduce solvent loss
Reduce internal shorting
Improve force output, actuation speed

21. IPMC Actuators

22. Results
Power [W]
Force [mN]

23. 100% Printable Robot

24. With Daniel Cohen, Larry Bonassar
Multi-material 3D Printer
CAT Scan
Direct 3D Print after 20 min.
Sterile Cartridge
Printed Agarose Meniscus
Cell Impregnated Alginate Hydrogel
Multicell print

25. The potential of RP Physical model in hours Small batch manufacturing
New design space
Design, make, deliver and consume products
Freedom to create

26. Learning from the history
Similarity with the computer industry
In the ’50s-’60s computers…
Cost hundreds of thousands of $
Had the size of a refrigerator
Took hours to complete a single job
Required trained personal to operate
Were fragile and difficult to maintain
Vicious circle
Niche applications  Small demand
Small demand  High cost  Niche applications
Digital PDP-11, 1969
Stratasys Vantage, 2005

27. Exponential Growth Source: Wohlers Associates, 2004 report
RP Machine Sales
Source: Wohlers Associates, 2004 report

28. The Killer App?
Honeywell’s “kitchen Computer”

30. Robust
Low cost
Hackable

31. Fab@Home Precision: 25µm Payload: 2Kg Acceleration: 2g
Volume: 12”x12”x10”

33. Fab@Home: “Fablab in a box”

35. Digital Structures

36. Reconfigurable systems
Fukuda et al: CEBOT, 1988
Yim et al: PolyBot, 2000
Chiang and Chirikjian, 1993
Rus et al, 1998, 2001
Murata et al: Fracta, 1994
Murata et al, 2000
Jørgensen et al: ATRON, 2004
Støy et al: CONRO, 1999
Zykov, Mytilianos, Adams, Lipson Nature (2005)

38. Programmable Self Assembly
Stochastic Systems: scale in size, limited complexity
Whitesides et al, 1998
Winfree et al, 1998
[Josh]: ok guys, before we jump into the hardships of hardware, lets see what this system might look like in simulation.
[tell the story of reconfiguration as movie plays].
As you can see, we’re not just interested in self-seembly – we’re interested in dynamic, programmable, reconfiguration.
There are many interesting algorithmic challenges, and we explore some of these in the paper: Key results are…
So Paul, did we get any of this to work in hardware? …

39. Saul Griffith, Nature 2005

40. Hardware implementation: 2D
[Paul:] ***In 2004, we showed two stochastic self reconfigurable systems in 2D: Each module had three degrees of freedom, and the structure assembled into a 2D pattern and then reconfigured using only “Brownian motion”. We used one system with swiveling permanent magnets, and one ‘solid state’ system with electromagnets. Kelvins in University of Washington recently achieved formation of patterns using 6 unit, and is also exploring how graph grammars can be physically implemented in such systems.
Solid state implementation (no moving parts) are important for scalability into microscale.
So Paul, what would be a natural next step to take?
[Paul:] ***Well Viktor, here we will explore reconfigurable self assembly of systems with 6 Degrees of Freedom. And more importantly, we seek a method of self assembly and reconfiguration that scales well to the microscale. I believe Daniela Rus is also working in this direction.
White, Kopanski & Lipson, ICRA 2004

41. Implementation 1: Magnetic Bonding
[Paul]: Two different sets of experiments were run to demonstrate self-reconfiguration and to gather statistics about the mean time to bond.
Start the movie
In the first experiment, one module is placed on the substrate while the other is placed at an initial condition that allows for repeatability in experimentation. The experiment begins by supplying power and agitation to the system.
When the module on the substrate powers on, it detects that it is the first module in the structure and waits for a dormant module to attach to it. When another modules attaches to first module, both modules detect one another and the module on the substrate determines that it has completed one configuration. Using serial communication, the module on the substrate coordinates a rejection sequence with other module. At the same instant, both modules pulse their electromagnets at the same polarity and the second module returns to float about in the medium.
Slowly, the dormant module is carried by the energy of the environment until it comes into a range where the magnetic force of the first module’s side causes the dormant module to attach. The experiment ends with the modules sitting in this reconfigured state.
With Paul White, Victor Zykov

42. Construction Sequence
High Pressure
Low Pressure

43. Construction Sequence

44. Construction Sequence

45. Construction Sequence

46. Construction Sequence

47. Construction Sequence

48. Reconfiguration Sequence

49. Reconfiguration Sequence

50. Implementation 2: Fluidic Bonding
[Viktor]: See for yourself.
[Paul]: Wait, but this isn’t in the paper!
[Viktor]: No, like any experimental work, we got it to work only a day before the conference.
Throughout this demonstration, the central cube is attached to the substrate. It is programmed to attract the second freely floating module to its right-hand side, then release it and re-attract it to its upper side. This movie has been accelerated 16 times. In real time, the experiment completes in 10 minutes.
[Paul]: Excuse me, how could you achieve 3D reconfiguration with only 2 cubes?
[Viktor]: Thanks - Paul, that’s a good point.
Accelerated x16
Real Time
With Paul White, Victor Zykov

51. 500 µm
With David Erickson, Mike Tolley

52. Tile dimension: 500μm
With Mike Tolley, Davis Erickson

53. Cytoskeleton of a mammalian cell
Randomized Machines
Tensegrity Robotics
Particle Robotics
Dictyostelium
Don Ingber, Scientific American 1998
With Chandana Paul
Cytoskeleton of a mammalian cell

54. Grand Challenges
Can we design machines that can design other machines?
Can we make machines that can make other machines?
Can we make machines that can explain other machines?
I’d like to conclude my talk today with what I consider to be two “grand challenges” of engineering.
First, how can we design machines that can design other machines? And second, how can we make machines that can make other machines?
Why are these challenges important?
First, these two abilities are a uniting theme of engineering, and automating these processes will require us to address, in a very quantitative and algorithmic way, the underlying principles of our discipline.
Second, making progress on these fronts will have huge leverage in many areas. Modern products are becoming increasingly complex, their complete design and fabrication often beyond the scope of understanding of any single person. Automating these processes will be an inevitable part of sustaining future progress.
Finally, I’d like to say that Biological life has manages to address these two questions in ways that dwarf the best teams of engineers; It is therefore worth our while to look at nature: Not just to mimic the final results, but to truly understand the adaptive processes that led to their emergence, and harness these principles to our benefit.
Thank you.