A Planning Framework for Robotic Insertion Tasks via Hydroelastic Contact Model

Robotic contact-rich insertion tasks present a significant challenge for motion planning due to the complex force interaction between robots and objects. Although many learning-based methods have shown success in contact tasks, most need sampling or exploring gather sufficient experimental data. However, it is both time-consuming expensive conduct real-world experiments repeatedly. On other hand, while virtual world enables low cost fast computations by simulators, there still exists huge sim-to-real gap inaccurate point model. finite element analysis might generate accurate results computationally expensive. As such, this study proposes framework with bilevel optimization leverage relatively information computation time. This consists of Dynamic Movement Primitives (DMPs) used parameterize trajectories, Black-Box Optimization (BBO), derivative-free approach, integrated improve policy hydroelastic model, simulated variability account visual uncertainty real world. The accuracy model then validated comparing our benchmark Peg-in-Hole task. Using these DMPs BBO trajectory generated capable guiding robot towards successful iterative refinement.