A Model-Based Approach for Dynamic Modelling and Collision Detection of Robotic Manipulators

The presented dissertation describes a workflow for the modelling, identification, and experimental validation of a robotic manipulator, with the aim of testing its capability to detect external collision. Starting with the theoretical formulation, the dynamic equations of a 2-DOF manipulator were derived using the Euler-Lagrange method and validated through numerical simulations in Simscape. Later, the study was extended to more complex structures, including the Parallelogram Arm and the PRRR manipulator, addressing the challenges of closed-chain kinematics through an energy-based generalization. To estimate the inertial parameters of the manipulator dynamic model, an algorithm for system identification was developed. Several estimation algorithms were implemented and compared. Among these, the WLS method, combined with parameter reduction based on relative standard deviation, provided the most accurate and stable results. Moreover, a data processing pipeline was designed to filter and differentiate the acquired signals, while optimal trajectories were generated based on the principle of persistence of excitation. This has led to the adoption of a ninth-order sinusoidal trajectory for the identification phase. The validation confirmed that the identified parameters generalized effectively to new trajectories. Finally, the dynamic model identified was applied to the problem of collision detection. Experimental tests were performed under controlled impact conditions. These have demonstrated that the model is capable of successfully discriminating between nominal operation and external contacts through the analysis of residuals among measured and estimated joints torques. The results obtained confirm that a dynamic physics-based model, when properly identified, can serve not only as a predictive tool for torque estimation, but also as a reliable means for collision detection and robot monitoring during interactions with the environment.