Search papers, labs, and topics across Lattice.
This paper develops a robust operational space control framework that integrates an extended state observer (ESO) with conformal prediction to enhance safety and adaptability in redundant robotic manipulators operating in dynamic environments. By estimating disturbances directly in operational space without relying on full-state measurements, the proposed method overcomes the limitations of classical computed torque controllers and data-driven residual learning approaches. Experimental results on a 7-DoF Franka Research 3 manipulator show millimeter-level tracking accuracy and real-time safe control at 1 kHz, even in the presence of significant disturbances.
Achieving millimeter-level tracking accuracy and real-time safety guarantees in robotic manipulation, even under dynamic uncertainties, could redefine standards for safe human-robot interaction.
Redundant robotic manipulators operating in constrained and human-interactive environments require accurate task-space tracking together with rigorous safety guarantees under dynamic uncertainties. Classical operational space computed torque controller (OSCTC) relies on accurate dynamic models and degrades in the presence of disturbances. In contrast, the data-driven paradigm of residual learning approximates disturbances as functions learned from full-state measurements, which are often noisy in practice, lack rigorous theoretical guarantees, and introduce additional design complexity. This paper proposes a robust OSCTC framework that integrates an extended state observer (ESO) with conformal prediction to combine model-based robustness and data-driven adaptability. The ESO estimates lumped disturbances directly in operational space without requiring full-state measurements as in residual learning, and a robust control barrier function (CBF) is constructed to enforce safety under uncertainty. However, robust CBFs require a known disturbance-variation bound to guarantee absolute safety, which often leads to conservatism in practice. To address this limitation, we further employ a sliding-window conformal prediction mechanism to estimate the bound online in a distribution-free manner, thereby achieving practical probabilistic safety guarantees. Experiments on a 7-DoF Franka Research 3 manipulator demonstrate millimeter-level tracking accuracy and real-time safe control at 1~kHz under various disturbances.
