This paper investigates the behavior of causal Jacobi differentiator in online robot identification applications. The identification process is carried out online using robot dynamic model. Causal Jacobi differentiator is applied to compute the acceleration from the noisy position measurements, with an exact time delay. Moreover its frequency domain property is analyzed via a finite impulse response (FIR) filter point of view, indicating clearly the differentiator performance. In the end, a two revolute joints planar robot identification simulation is presented with adaptive controller.