Robot trajectory prediction relies on real-time, accurate sensory data and reliable representation models. Moreover, the handling of uncertainty during environment interaction fundamentally determines its accuracy. However, existing prediction models cannot meet the requirements of robot trajectory characterization, and data-driven prediction models fail to satisfy the interpretability demands of trajectory prediction. On this basis, this paper proposes a trajectory characterization method using kinematic equations guided by control mechanisms, employing mechanism-embedded modeling to enhance interpretability and interactivity. To respond the dynamic updating requirements of temporal model parameters, the parameter estimation process is firstly treated as a Markov decision process (MDP). A trial-and-error estimation method with the maximum likelihood function reward is constructed to realize the dynamic estimation and updating of the model parameters. Based on the robotic trajectory characterization with control mechanism-guided reinforcement learning, the methods of trajectory tracking and prediction are proposed respectively. Finally, the effectiveness of the model is verified by eight sets of simulation cases.