Scaling Trusted Autonomy: Towards Generally Capable Verifiable Robots

This research paper argues that scaling trusted autonomy requires a neurosymbolic approach unifying learned models with symbolic structure. It introduces three complementary directions: LIMP, which uses foundation models and spatial reasoning to ground natural-language instructions in explicit task constraints; REALM, which maintains an editable photorealistic 3D Gaussian scene to predict manipulation consequences; and PACTS, which learns visuomotor skills that jointly generate action trajectories and predictions of their symbolic effects. Together, these contributions aim to provide a foundation for generally capable robots whose behavior can be assessed under the assumptions of their models.