Dynamic Physical Rendering

1
Dynamic Physical Rendering
What if software could morph physical shapes in
real time?

• New ways to do medical visualization
• Better interpret 3D data (CAT, MRI)
• Rehearse procedures off-patient
• Control surgical robots

• New approaches to design
• Interact with CAD results immediately
• Reshape a physical object (like molding clay),
  see the results flow back to CAD

• Reconfigurable antennas
• - Programmable antenna, reflector shapes
  complement software defined radio

• Games, entertainment, home movies

• Shape-shifting handhelds
• Convenience
• Optimize user interface for the task at hand
• Create new form-factors with software alone

• Be there without being there
• Not like videoconferencing-jail
• Full interaction doctors make house calls,
  tennis instructors demonstrate technique

How might such programmable matter be
constructed?
Each 3D shape reproduced is formed of millions of
tiny cooperating spherical modules
one module(catom)
Moving 3D reproduction, using millions of tiny
spherical modules
Moving 3D Capture

• Each module contains
• processing, networking, energy storage, etc.
• a means of actuation (locomotion and adhesion)
• outer surface of each module is a video display

• Properties of programmable matter
• Not just the an illusion of 3D (as with stereo
  glasses), but real, tangible physical objects
• Both an output device (rendering) and an input
  device (haptics, sensing)
• Each robot is a physical voxel

2
What fundamental computer systems problems need
to be addressed?
1. How do we cope with scale dynamism on these
levels? 2. How can we create self-organizing
infrastructure for power, networking, physical
support, proprioception, debugging, etc.? 3. How
should we arrange and operate myriad small-scale
actuators such that the ensemble can achieve
large-scale forces and motions?
Current Research Efforts

• Power Routing
• Shape Formation
• Programming Languages for Emergent Behaviors
• Macro-Scale Motion Planning
• Collective Actuation and Dynamic Motion
• Hardware Prototypes
• Distributed Consensus and Control
• Debugging Tools for Massively Distributed
  Software
• Adaptive Hierarchies for Communication and
  Control
• Scalable Coordination
• Scalable Physics-based Simulation

Collaborative Research
Student Alumni
Current CMU Students
Carnegie Mellon
Intel
Burak Aksak Nels Beckman Preethi Srinivas Bhat
Mike De Rosa Daniel Dewey Stanislav Funiak
Emre Karagozler Brian Kirby Eugene Marinelli Ram
Ravichandran Ben Rister Michael
Weller Byung Woo Yoon
Jason Campbell, PI Phil Gibbons Casey
Helfrich Todd Mowry Lily Mummert Padmanabhan
Pillai Siddhartha Srinivasa Rahul Sukthankar
Khalid El-Arini (CMU) Greg Reshko (CMU) Lauren
Chikofsky (CMU)Ashish Gupta
(Northwestern) Bancha Dhammarungruang
(CMU)
Seth Copen Goldstein, PI Johnathan Aldrich Gary
Fedder Carlos Guestrin James Hoburg James
Kuffner Peter Lee Matt Mason William
Messner Illah Nourbakhsh Metin Sitti Srini
Srinivasan Dan Stancil Manuela Veloso
Elsewhere
Kasper Stoy (Univ. Southern Denmark) Mark Yim
(Univ. Pennsylvania) Ashsih Deshpande
(student, Univ. Michigan)
Funding
Intel DARPA National Science Foundation Carnegie
Mellon University
Intel Research Pittsburgh Interns
3
How can we render and actuate shapes using
millions of tiny, cooperating robots?
Stochastic, distributed shape formation with Hole
Motion

• Create and delete holes to modify surface
  contours
• Holes move like gas molecules
• Local rules automate hole movement
• Shape planning complexity is independent of
  number of modules

Exact, centralized shape planning with Hierarchies

• Optimal planning scales exponentially
• Instead, group modules into self similar
  hierarchies of metamodules
• Recursively apply precomputed templates to
  simplify planning

Continuous, forceful shape transformation with
Collective Actuation

• Parallel movement of modules to effect
  larger-scale motion
• Can create flexible, extensible structures
• Can collectively exert large forces against
  external objects
• Continuous, smooth global shape changes