How do we make robots actually reliable in the real world?

Robots deployed in open environments face reliability challenges from uncertainty, human interaction, and emergent behaviors across coupled components. The authors outline three complementary approaches: scenario-based testing with validated specs, compositional verification using symbolic representations, and runtime assurance that adapts to distribution shifts. The key insight: these must integrate through shared neuro-symbolic representations and continuous lifecycle feedback, not operate in isolation.