What this talk is about
Origin coordinates a body-area network of energy-harvesting sensors — wrist, ankle, chest — for human activity recognition. The talk frames the problem (each node is independently intermittent, so the system has to coordinate them without a central always-on clock) and walks through two key ideas:
- Activity-aware scheduling. Decide which sensor attempts the next inference based on (a) the energy the sensor expects to harvest in the next window and (b) which sensor’s vantage point is most useful for the predicted activity class.
- Lightweight ensemble aggregation. Fuse partial and confident results from whichever nodes successfully completed their slice. The ensemble is structurally tolerant to missing voters — losing one sensor to a bad harvest window degrades gracefully.
Across standard HAR datasets, this nets 2.5–5% higher classification accuracy under sporadic harvested power, with negligible coordination overhead. Origin was a Best Paper Nominee at DATE ‘21.
Quick references
- Paper: Publications page → Origin (DATE ‘21)
- Project deep-dive: /projects/origin/