Predictive Homeostasis — A Self-Modeling Organism

Environment

Disturbance amplitude 35

Sensor noise R 8

Controller gain K_p 0.55

Actions

Self-Model Readout

True state x

50.0

Estimate x̂

50.0

Prediction error

0.0

Kalman gain K

0.00

Organism health 100%

Homeostasis stable

How it works

The organism never observes its true internal state x directly — only a noisy sensor reading z. It keeps a running belief x̂ (its "self-model") and updates it every tick with a scalar Kalman filter:

predict:  x̂⁻ = x̂ + u·dt        (u = last self-correction)
          P⁻ = P + Q             (belief uncertainty grows)
update:   K  = P⁻ / (P⁻ + R)   (Kalman gain, 0..1)
          x̂  = x̂⁻ + K·(z − x̂⁻)  (blend prediction with sensor)
          P  = (1 − K)·P⁻

K automatically weighs trust between its own prediction and the incoming sensor: a noisy sensor (high R) pushes K toward 0 (trust the model), a confident sensor pushes K toward 1 (trust the reading).

The organism then acts on its own belief, not on reality — classic biological control:

u = −K_p · (x̂ − setpoint)      (corrective action)
x_true += dt·(u + disturbance + process noise)

Disturbance amplitude — how hard the environment pushes the true state away from setpoint (feeding, temperature swings, external stress).
Sensor noise R — how unreliable the organism's own sensing is; push it high and watch the Kalman gain drop, so the self-model leans on prediction over noisy input.
Controller gain K_p — how aggressively the organism corrects toward its setpoint once it believes it has drifted; too low and it drifts, too high and it overshoots and oscillates.
Inject Shock — a one-off large disturbance, testing how fast the feedback loop recovers.

This is the same predict → sense → correct loop behind thermoregulation, blood-glucose control, and — in engineering — every Kalman-filter-based autopilot or sensor-fusion stack; "complete knowledge of its own state" is really just a continuously-updated probabilistic belief, never a perfect measurement.

True internal state Self-model estimate

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.