We introduce Wearable Mishap Avoidance (WMA), in which always-on AI for a wearable device detects impending mishaps in an imperfect legacy system (human or cyber-physical) and issues a corrective alert before it is too late. The time-critical sensing and actuation loop of this new class of applications involves a tradeoff betweeen rapid reaction (to meet externally-imposed mishap deadlines) and patience (to allow the legacy system to self-correct in time). We show that WMA can be implemented today using off-the-shelf wearable hardware, wireless networking, edge computing, and AI models. We present a model for reasoning about WMA, implement proofs of concept for three different use cases, and discuss future research directions.

25 pages

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu