Morphological Computationalism
From the AI Research Compass wiki
Morphological Computationalism: A well-designed body does much of the computing itself, so intelligence should be offloaded into materials, springs and shape.
Beliefs
Morphological Computationalism argues that the brain is not the only computer in a robot. A passive walker strolls down a slope with no controller at all; a soft gripper conforms to an egg without a grasp planner; a whisker’s stiffness already filters the signal before any neuron sees it. Choosing the right shape, compliance and material turns hard control problems into easy ones, and sometimes into no problem at all. Intelligence, on this view, is distributed between body, brain and environment, and the cheapest computation is the one physics does for free.
It sits at the far World edge and low toward the Harness end. It is maximally physical, but it is not a giant end-to-end policy: it is about co-design, with intelligence spread across body, controller and environment, which pulls it down the vertical axis. Its shade is Beyond Revolutionary because mainstream robot learning still treats the body as a fixed given and pours all the cleverness into the network.
Arguments
- Passive dynamic walkers demonstrate stable, efficient gait with little or no active control, showing the body can carry the computation.
- Compliant and soft grippers handle uncertainty in object shape and pose that rigid hands need elaborate perception and planning to manage.
- Offloading work into mechanics reduces the sensing, compute and data a controller needs, which matters for small, cheap or energy-limited robots.
Critiques
- Morphology is hard to change once built, while software can be updated daily; the doctrine trades flexibility for elegance.
- Bodies tuned for one task can make other tasks harder, so morphological solutions may generalize poorly.
- Large learned policies increasingly cope with ordinary rigid hardware, weakening the claim that special bodies are necessary.
Representative works
Relationships
Friends
Frenemies
How to get it
Land on this cell in the full or pro test: x ≈ +0.95, y ≈ −0.75, paradigm ≈ +0.90.