+physical devices for sensing, interaction, and utility — a cluster that reflects a consistent pull toward hands-on building at the hardware/software boundary. the ideas span from human-computer interaction (hands-free control via [[emg-bracelet|EMG bracelet]] or [[pupilometry-glasses|pupilometry glasses]]) to signal sensing ([[sensor-capturer|sensory capturer]], [[ppg-biomarker-wearable|PPG biomarker wearable]], [[eeg-artifact-rejection|EEG artifact rejection]]) to environmental detection ([[acoustic-drone-detection|acoustic drone detection]], [[predictive-maintenance-sensors|predictive maintenance sensors]]).

+the highest-rated ideas in this cluster are the ML-on-hardware ones: acoustic drone detection, PPG biomarker wearable, and predictive maintenance sensors all scored [STRONG] on the spreadsheet because they combine embedded systems engineering with meaningful ML research. the common pattern: novel sensor signal → custom dataset → trained model → edge deployment. this is technically demanding in an interesting way — TinyML, quantization, model compression — and produces demos that are viscerally impressive. [[eeg-artifact-rejection|EEG artifact rejection]] is the most academically serious of these: self-supervised neural signal cleaning is a real open problem with research significance.

+

+the human-computer interaction ideas ([[emg-bracelet|EMG bracelet]], [[pupilometry-glasses|pupilometry glasses]], [[robotic-arm-assistant|robotic arm assistant]]) target hands-free computing from different angles. the bracelet approach is probably most practical given that Meta is working on the same thing with smart glasses wristbands, validating the space. [[sensor-capturer|sensory capturer]] and [[nose-device|nose device]] are the most exploratory — capturing sensory experiences beyond audio/video is a genuinely underdeveloped area that connects to the [[cluster-memory-and-context|memory and context]] cluster, where richer sensory capture enables richer memory recall.