Deadline-Aware Runtime Assurance for Autonomous Mobile Robots Under Sensor Degradation
Abstract
This research monograph develops a joint sensing-and-execution recoverability formulation for autonomous mobile robots. A conservative braking supervisor combines bounded localisation uncertainty, measurement age, outstanding commands and computation delays to determine an admissible speed and intervention budget. The evaluation comprises 100,000 algebraic boundary states, 32,400 main planar navigation episodes, 3,900 supplementary navigation episodes, 24,000 integrated wall approaches including beyond-bound tests, and four independent-process software-in-the-loop timing cases. The monograph contains 11 chapters, 13 figures, 13 tables and 19 references. The joint certificate produced zero false-safe decisions in the sampled bounded boundary experiment and zero collisions in 3,000 bounded wall approaches. Violating braking and response-delay bounds produced failures. On the six static navigation maps, joint supervision completed 5,268 of 5,400 episodes, compared with 2,674 for a global worst-case supervisor and 5,400 for nominal navigation. All controllers recorded zero sampled collisions on those navigation routes. Learned anticipation reduced held-out prediction error but left held-out task completion unchanged. The experiments use a reduced-order robot model, bounded Cartesian position observations, static obstacles and process-based execution measurements. The discussion states the modelling assumptions, benchmark interpretation and extensions to richer recovery manoeuvres and robotics middleware. Revision 02 streamlines the front matter and editorial presentation. Mathematical methods, recorded experimental data and numerical results are unchanged. This record contains the LaTeX-generated PDF.
