Morphological Anarchism

From the AI Research Compass wiki

Morphological Anarchism: Bodies and brains should co-evolve freely; no fixed robot form, humanoid or otherwise, deserves to be the default.

Beliefs

Morphological Anarchism rejects the assumption that we already know what a robot should look like. Engineers design a body first and then train a controller for it, but in nature body and nervous system evolved together, and much of the “intelligence” of a gecko or an octopus is in its shape and materials. So it lets morphology itself be searched: limb counts, joint placements, stiffness and sensor layout are all variables, optimised jointly with the policy. The result is often strange, sometimes a wobbling blob, and that is the point.

It sits at the far World edge, where the body is not a peripheral but part of the computation, and near the Harness end, because there is no single central model governing everything; control is spread across a body whose physics does part of the work. The Beyond Revolutionary shade comes from refusing both the fixed-form robot and the humanoid consensus at once.

Arguments

  • Classic work on evolved virtual creatures showed that co-evolving body and controller produces effective, often surprising locomotion strategies no designer specified.
  • A well-chosen morphology can make control trivially simple, as passive dynamic walkers demonstrate, so searching over bodies can save more than searching over policies.
  • Fixing the body to a human template imports human limitations into tasks that could be done better by other shapes.

Critiques

  • Evolved morphologies are usually tested only in simulation, and exploiting simulator quirks is a notorious failure mode.
  • Manufacturing a new body for every task is impractical; standard forms exist because tooling, repair and data sharing depend on them.
  • The search space of bodies times policies is enormous, and most results remain small-scale demonstrations.

Representative works

Relationships

How to get it

Land on this cell in the full or pro test: x ≈ +1.00, y ≈ −0.75, paradigm ≈ +1.00.

A first draft. Corrections and better arguments are welcome through a pull request. Edit this entry on GitHub Edit its data