An Adaptive Longitudinal Platooning Design Based On Concurrent Learning

What the math gives to ML

The paper's transferable mechanism is concurrent learning: parameter adaptation uses both current and replayed data, so convergence can be obtained without persistence of excitation. For a scalar unknown decay parameter, one stored sample with a nonzero regressor creates a positive information lower bound and yields exponential parameter-error decay, even when subsequent online data are uninformative. A direct neural-network application is online calibration of leak or decay rates in state-space and recurrent layers, combined with projection constraints that preserve contractive hidden-state dynamics.