Blog topic
How movement becomes measurable, interpretable signal.
10 stories · Latest Oct 14, 2026
A population reference says where a person sits among strangers; a personal baseline says whether they have changed. What a baseline is made of, how many observations it needs, when it moves, and what it can and cannot say.
Mobility8 min read
Read the storyA model can be right on every frame while the system around it fails: dropped frames, stalled inference, a network that swallows an alert, an interface that hides it, a person who has stopped trusting it. Model quality is one link in the chain.
Engineering8 min read
Read the storyRGB, face removed, silhouette, skeleton, joint trajectories: each representation removes some identifying cues and keeps others. Where identity disappears depends on storage, combination, persistence and purpose — not on the picture alone.
Responsible AI8 min read
Read the storyA camera records a projection of a walk, not the walk. The angle decides which signals survive the flattening — knee flexion, stride width, timing, foot trajectory, body path — and a movement system has to be honest about which it has.
Engineering9 min read
Read the storySymmetry compares the left side of a walk with the right — but stance, swing and step timing can each be asymmetric on their own, nobody is perfectly symmetric, and what a change means depends on the baseline it is read against.
Movement intelligence8 min read
Read the storyPose estimators almost always return a plausible skeleton. The failures that matter are the ones that still look like a person: a misplaced ankle, crossed legs, a foot that vanishes — and every gait measure downstream inherits them.
Movement intelligence10 min read
Read the storyA walking video looks simple. Turning it into reliable movement intelligence is not. Inside the pipeline from capture and pose estimation to gait features, sensor fusion and actionable outputs.
Movement intelligence8 min read
Read the storyGait has often been studied as a biometric signature. But human movement can carry information about mobility, recovery, functional change, risk and safety context as well.
Movement intelligence4 min read
Read the storyMany movement-intelligence tasks do not inherently require identity. A privacy-aware architecture starts by asking what information the task actually needs.
Responsible AI5 min read
Read the storyA single assessment tells us how someone moves today. Longitudinal movement analysis asks the more useful question: how is that movement changing?
Mobility4 min read
Read the story