Robot Federalism
From the AI Research Compass wiki
Robot Federalism: A robot should be a federation of separate perception, planning and control modules with clean interfaces, not one end-to-end network.
Beliefs
Robot Federalism is the classic architecture of robotics: divide and rule. Perception builds a map and tracks objects; planning searches for paths and task sequences; control turns trajectories into motor commands; a middleware layer passes messages between them. Each module is owned by specialists, can be tested in isolation and swapped for a better one. The federal constitution is the message interface, and a well-defined interface is worth more than a clever monolith nobody can debug.
It sits well on the World side and at the very bottom, the Harness extreme: firmly physical, and organized as a system of components. Its neutral shade marks an established tradition that built most deployed robots, from warehouses to research labs, without betting on either the scaling recipe or a new paradigm. Its rivals are the robot foundation-model camps that want one policy for every body and task; federalists welcome learned modules inside the federation but resist handing the whole government to a single network.
Arguments
- Modular robot stacks are the basis of most deployed robots, and their components can be validated and certified separately.
- Shared middleware and message standards let research groups reuse each other’s perception, planning and control modules.
- Failures can be traced to a specific module, which makes debugging and safety analysis tractable.
Critiques
- Hand-designed interfaces force lossy intermediate representations, such as maps and object lists, that may not capture what the task needs.
- Integration overhead is large: getting many modules, timings and coordinate frames to agree is a notorious time sink.
- End-to-end and foundation-model camps argue that federations generalize poorly to unstructured tasks that cross module boundaries.
Representative works
Relationships
Friends
How to get it
Land on this cell in the full or pro test: x ≈ +0.51, y ≈ −1.00, paradigm ≈ +0.05.