PIDNet controller slashes tracking error 10% on thrust vectoring platform

A new paper on arXiv (2607.27793) presents PIDNet, a robust attitude-tracking controller for dual-actuator thrust vectoring platforms. The system uses two linear actuators connected through a universal joint to orient a thrust direction on the unit sphere. The core idea is to pair a bounded nonlinear proportional-derivative (PD) term with a radial basis function (RBF) network that performs online weight adaptation, generating state-dependent integral compensation using only tracking error and its derivative from actuator-displacement measurements. This model-free, coupled MIMO structure lets the RBF network naturally compensate for cross-channel effects like direction-dependent friction and geometric coupling.

Stability is proven via a convex Lyapunov function with a bounded gradient, and the authors derive an H-infinity gain bound for the extended error. Experimental validation on a two-degree-of-freedom gimbal rig shows PIDNet achieves better tracking performance than both a conventional PID and a super-twisting sliding-mode controller. The online adaptation reduces the integral of time-weighted error (ITNE) by 10% compared to the non-adaptive baseline, with no increase in control effort. This makes PIDNet a practical, high-performance option for aerospace, robotics, and any system requiring precise thrust vectoring or attitude control in the presence of uncertain dynamics.