Learning from Demonstration (LfD) is a key methodology for enabling robotic autonomy, allowing systems to acquire new skills by imitating human demonstrations rather than relying on low-level programming. In surgical robotics, LfD supports the modeling, generalization, and reproduction of expert trajectories. This work presents a multidimensional LfD framework that extends the classical Douglas–Peucker algorithm to integrate both kinematic and dynamic (force/torque) data, enabling effective simplification and encoding of trajectories in multidimensional spaces. This enriched representation enhances both interpretability and fidelity by capturing motion geometry alongside interaction dynamics. The encoded data are used to train a Hidden Markov Model (HMM), which generalizes motion patterns from multiple demonstrations. The framework was validated experimentally using a dataset of expert-performed needle insertion trajectories. Results show that the model accurately reproduces representative motion patterns under varying simplification and discretization conditions, while all generated trajectories remain within clinically acceptable deviations. These findings demonstrate that the proposed multidimensional LfD model offers a robust and scalable solution for encoding, learning, and generating complex surgical trajectories with improved adaptability and reliability.