Physics-based reactive motion planning for robot manipulators: design, simulation, and experimental validation.

This thesis proposes a motion planning and control method based on Virtual Model Control (VMC), called the Multi-Segment Virtual Mechanism (MSVM) and developed for the Franka Research 3 (FR3) manipulator. The method comprises three components, all implemented as passive virtual mechanisms and integrated into a single dynamic solver: a planner (MSVM), whose trajectory coincides with the minimum potential energy configuration of a chain of telescopic segments, a multi-segment virtual slider, which compliantly guides the end-effector along the path, and a Whole-Body Avoidance strategy, which protects the entire kinematic structure. Since the path is the minimum of a potential, it is continuous and smooth by construction, without the need for subsequent optimisation steps. The distinctive contribution is the elimination of re-planning: moving obstacles are handled by the continuous control loop at 1 kHz, which reactively deforms the virtual mechanism rather than recalculating the path. The passivity of the entire closed-loop system also guarantees stability, ensuring robustness to model errors and a reliable simto- real transfer. The method has been validated in simulation and on the real robot. The planner identifies a collision-free trajectory in 100% of trials with increasing obstacle density. The execution success rate reaches 97.5% in a static environment and 99.5% in a dynamic environment, with planning times of less than one second. On the FR3 robot, the method was tested in four scenarios, handling dynamic obstacles with peak speeds of up to 1 m/s.