Robotics systems traditionally treat actuation and sensing as direct extensions of computation. Controllers generate commands, middleware transmits them, and actuators execute. Authority is positional: if code runs within the control stack, its commands are presumed legitimate. Safety mechanisms are layered on top, but the ontology of actuation remains unchanged. Constitutional Robotics proposes a foundational programming model reform. All reality-touching operations - actuation, sensing, persistence, communication, and time - are expressed as constitutional system calls. Instead of invoking hardware directly, controllers submit proposals to a governed boundary. An explicit admissibility function evaluates authority, policy, and safety constraints before returning a structured outcome: execute, simulate, defer, or refuse. This architectural shift separates computation from jurisdiction. The computational core remains effect-pure; world interaction becomes an adjudicated boundary crossing. The paper formalizes: A minimal robotics syscall interface Compositional multi-authority evaluation chains Structured refusal semantics Legitimacy-state feedback as a first-class control signal A constitutional logging substrate enabling deterministic replay and forensic auditability Constitutional robotics unifies hardware and simulation pathways under a shared syscall boundary, enabling replay parity between physical systems and digital twins. Refusal becomes structured information rather than implicit override. Controllers gain visibility into admissibility constraints, enabling authority-aware adaptation and safer planning. Rather than treating robots as unilateral actors, constitutional robotics reframes them as petitioners operating under explicit contracts. The result is a drift-resistant, auditable, and jurisdictionally coherent foundation for autonomous systems operating in safety-critical and regulated environments.
Adam Ableman Mazurk (Mon,) studied this question.