Sunday, November 10, 2013

Morphs: Moving Architecture with Decentralized Brain


Architect William Bondin calls his structures Morphs, an acronym for 'Mobile Reconfigurable Polyhedra.' These are intended to move slowly through public spaces such as parks and engage and interact with the people present there. The core innovation here is that the structure is built out of tetrahedras which consist of trusses that can extend to up to twice their length. This property allows a tetrahedra to shift its center of gravity so that at some point it can flip over in any of the three directions. This allows the structure to navigate through a space.

To do this navigation, the system relies on some simple principles that are in accordance with embodied cognition. It has sensors distributed throughout the system that measure properties such as the presence of water, temperature, and light, and based on simple rules it learns to behave so that it will avoid water, roads and shaded areas, and go towards areas more crowded with people and with more sunlight. It also learns about its environment in order to determine where it has already been and where it should go. This behavior is inspired by slime molds, which offload their memory to the environment by depositing slime so that they know where they have already been.


The environment is a crucial factor in designing embodied agents, as Rolf Pfeifer and Josh Bongard describe in the book "How the Body Shapes the Way we Think". And it is often overlooked by designers of intelligent agents, who may tend to think from a cognitivist perspective and focus on the brain as the main controller of behavior. We need to think of behavior as a dynamically evolving interplay between the agent's brain/neural system, body and the environment. Together, these factors make the system reach certain 'attractor states' of behavior that fit the embodiment in an optimal way so that it will behave most efficiently. The book mentions some interesting ideas about how to have the environment contribute to the emergence of intelligent behavior:

- 'Scaffolding': Structuring the environment with passive objects to simplify tasks. With adequate scaffolding, mechanisms for successful navigation will be very cheap.
- Offloading memory to the environment - information about the past can be contained in the current situation this way and included directly into sensory-motor loops rather than requiring cognitive action.
- Getting the material properties (elasticity, stiffness, damping etc.) of the body exactly right to fit the environment is crucial. If we get the material properties right, the desired behaviors ('trajectories towards attractor states') will emerge from the interaction with the environment. I can't stress enough that every detail is crucial here - imagine for example if your achilles was only a little bit less elastic or a little longer; it would completely change the way your entire body moves!
- Placement of sensors is also crucial - it is best to have sensors there where the most structured data can be derived from the environment.

Based on these kinds of principles for embodied intelligence, as the Morphs are a nice example of, I would like to see much more innovation!