Plan the trajectory, not only the next node.
The one-step selector remains valid only when configuration ordering stays unchanged throughout the required dwell interval and no candidate loses admissibility inside it. Otherwise the controller plans over the graph on a…
Plan the trajectory, not only the next node.Open section
Plan the trajectory, not only the next node.
The one-step selector remains valid only when configuration ordering stays unchanged throughout the required dwell interval and no candidate loses admissibility inside it. Otherwise the controller plans over the graph on a receding horizon.
The revised model changes architecture decisions.Open section
The revised model changes architecture decisions.
A model that no longer predicts the observable world is degraded even if the compute platform is healthy.
Deep degraded and safe-stop states should not automatically return to the switching set when they dominate the dwell-time bound.
The guarantees belong to the verified graph. Runtime policy is allowed to learn only inside that graph.
The mechanism may need millisecond response while the sequence itself is constrained to change much less often.
The previous model is retained as part of the evidence trail.
EXP‑01, EXP‑02 and EXP‑03 remain intact. The new formulation is presented as the consequence of what those experiments could establish — and what they could not.