Dual Control for Active Estimation and Path Planning in the Automation of Robotic Assembly Tasks

This work extends the principle of dual control for exploitation and exploration to robotic assembly tasks, where successful assembly requires both accurate state estimation and informed control actions to reduce uncertainty. The assembly task is formulated as an active inference problem in which the robot maintains a probabilistic belief of target location and executes actions that simultaneously reduce uncertainty (exploration) and drive the system towards alignment(exploitation). The framework is validated on an industrial UR10e robot for circular and square peg-in-hole insertions, representing distinct geometric contact conditions. To realize this, a physics based measurement model is derived that relates contact forces to relative peg-hole position to support Bayesian estimation of the target state. Performance is evaluated under two uncertainty regimes: 1) small initial error (low bias and uncertainty) and 2) large initial error (high bias and uncertainty), and compared against a single step model predictive control and an entropy-based path planning strategy. Experimental results demonstrate that the proposed approach achieves robust and reliable insertions across a wide range of initial conditions, highlighting the potential of the proposed work for robotic assembly.