One object, three noisy senses. Vision reads its shape, a tap reads its material and — through the closed shell — its fill level, a fingertip reads its texture. Each is expert at one thing and near-blind at the rest. Watch Bayesian fusion make a scene understanding sharper than any single sense — and watch what happens to fill when you take the sound away.
The object · ground truth
Senses in play
The world gets harder
Sample
Belief — each sense alone, then fused · gold outline = the truth
A Bayesian sensor-fusion model. Each sense's likelihoods are set from its real strengths, not learned from data — the point is the logic of why fusion wins, which is exactly what SonicSense and ManiWAV learn from contact audio. Runs entirely on your device. · Institute for Physical AI @ BMI