Jointly Optimize High‑Level Planning and Low‑Level Control Based on Contact‑State Recognition

Determine effective strategies to jointly optimize the high‑level planning module (which generates desired forces and moments from geometric and environmental constraints) and the low‑level controller (e.g., position‑based force or impedance control) conditioned on contact‑state recognition for robotic peg‑in‑hole assembly.

Background

Within contact model‑based approaches, the authors distinguish a high‑level planning module that produces desired forces/moments and a low‑level controller that executes actions. Many existing works emphasize contact state recognition and low‑level control while neglecting the high‑level planning module.

The authors note that how to best coordinate and optimize both modules together based on contact state recognition is not yet clear, indicating a need for methods that integrate recognition results into both planning and control design.