KALMAN : Derivative is slow about xcvario

[tilaktilak]

Vetat = [a da/dt];

• Only 1 measurement : 1 measurement noise
• Estimate measurement noise by : Measure static output of sensor, variance should be 1/6 * (range of static data). Need to check a simulation of a random variable with gaussian noise
• Update look like this : measure = H*X + sigma_measure => check if sigma measure is a variance/ecart-type or random variable.
• Predict : Xk+1 = A*Xk => [1 dt; 0 1] + P ? Not sure
• ak+1 = dt*(dak/dt) + ak + SIGMA1
• (dak+1/dt) = (dak/dt) + SIGMA 2
• SIGMA 1 = 0 because model is true theorically. SIGMA2 should be set to a value near max acceleration that the system can take ( equivalent to dt*(d^2a/dt^2) but as we do not measure acceleration, it's estimated by a random variable). It's also possible to estimate derivative only with a SIGMA2, if system is really dynamic (but could diverge if spark on measurement or loss of sensor). It's also possible to have (da/dt)*0.98 + SIGMA2 to stabilize filter ( loss of sensor or spark )

[tilaktilak]

Problem seems linked to loop timing : Should measure dt to process kalman even if loop is late.

[tilaktilak]

Closed by e690d9b