INRoL
Interactive & Networked Robotics Laboratory
Department of Mechanical & Aerospace Engineering
Seoul National University
INDUSTRIAL CONTROL
Precision Motion Control of Robotized Industrial Hydraulic Excavators via Data-Driven Model Inversion
This work proposes a novel precision motion control framework of robotized industrial hydraulic excavators via data-driven model inversion. To approximate the excavator dynamics including input delays and dead-zones, we construct a physics-inspired data-driven model with a modular structure. Then it is inverted in a modular fashion to obtain the data-driven model inversion that compensates for the excavator dynamics, which constitutes the control framework with a proportional control to enhance the robustness. The framework is experimentally validated with a commercial 38-ton class hydraulic excavator, achieving a precise control performance (i.e., RMS of the path following error under 2 [cm]) even in the presence of intense soil interactions. (RA-L 2022)
Expert-Emulating Excavation of Industrial Excavator
We propose a novel excavation trajectory planning algorithm, which can effectively emulate the demonstration from experts by data-driven approach. For this, we encode the expert excavation trajectory with dynamic movement primitives (DMP) with GMM shaping force to robustly preserve qualitative shape of the trajectory. We also real-time modulate the expert emulating (nominal) trajectory to prevent excessive build-up of excavation force by using the real-time estimation of the excavation force. This result has been successfully incorporated in the Concept-X demonstration (IROS 2020, Concept-X (Link, News))
Estimation and Control of Tandem Cold Mill Process
TCM (Tandem Cold Mill) is a large-scale industrial system that produces cold-rolled steel (relatively thin and high quality) by sequentially pressing the entering hot-rolled steel. In order to address the thickness hunting from varying hardness, we propose a novel framework to estimate strip-longitudinal hardness of the TCM (Tandem Cold Mill) process and its feedforward control, while fully incorporating the interconnected nature and sensing sparsity of the TCM process. Efficacy of the proposed estimation and control frameworks are validated with high-precision TCM process physics simulator that we previously developed. (IFAC 2020)
We develop virtual sensors for gasoline LP-EGR based on an optimal estimation based on a mixed physics-based/data-driven modeling, which is more accurate than physics-based modeling while also faster than data-driven modeling. Extending Unscented Kalman Filtering theory to the new nonlinear uncertainty structure of this mixed model, we optimally combine the models/sensors w/ their modeling/sensing uncertainties while properly addressing the MLP structure of the data-driven modeling. We also formally analyze the computation load of the proposed framework, and experimentally validate this optimal virtual sensors for the LP-EGR. (IFAC2020)
LP-EGR Estimation w/ Mixed Physics/MLP-Modeling
We propose a novel modeling and control framework for HCCI engines, which, by utilizing direct in-cylinder pressure sensing, can detect, and react to, the wide spectrum of combustion, thereby, allowing for prevention of, or even recovery from, partial burn; and transient control with incomplete combustion and misfire avoided. For this, we first develop cyclic control-oriented model of HCCI process, in which Arrhenius integral is completely elimintated by quantities based on the in-cylinder pressure sensing. We then propose a nonlinear control based on feedback linearization and scheduled-LQR. (ASME JDSMC2018, Joint work with AESL at SNU)
HCCI Engine Control
Automatic excavator is expected to hit the market soon, significantly improve productivity, fuel efficiency and also safety. One of the key challenges of any such automated systems is safety. We develop passivity-based control of electrohydraulic excavator, which, by limiting imparted energy to the environment, can automatically and gradually cease the excavation when unforeseen danger is impending. We also incorporate MCV, which complicates modeling and control design due to its triggering of fluid circuit switching. (Automatica 2019, Link)