Workshops, 2nd October (Mon)

1. Practical Methods for Real World Control Systems

Monday, October 2nd, 8:00 AM - 12:00 PM
Room: Cobalt

Organizer: Daniel Y. Abramovitch

ABSTRACT:
The proverbial “gap” between control theory and practice has been discussed since the 1960s, but it shows no signs of being any smaller today than it was back then. Despite this, the growing ubiquity of powerful and inexpensive computation platforms, of sensors, actuators, and small devices, the “Internet of Things”, of automated vehicles and quadcopter drones, means that there is an exploding application of control in the world. Any material that allows controls researchers to more readily apply their work and/or allows practitioners to improve their devices through best practices consistent with well understood theory, should be a good contribution to both the controls community and the users of control. This workshop is intended as a small but useful step in that direction.

2. Differentiable Programming for Modeling and Control of Dynamical Systems

Monday, October 2nd, 8:00 AM - 12:00 PM
Room: Lapis

Organizer: Jan Drgona, Aaron Tuor and Draguna Vrabie

ABSTRACT:
In recent years there has been an explosion of research on the intersection of machine learning and classical engineering domains. Machine learning is increasingly being used in the development of novel data-driven approaches for modeling and control of dynamical systems, traditionally dominated by physics-based models and scientific computing solvers. On the other hand, engineering and scientific computing principles are changing the machine learning landscape from purely black-box into domain-aware methods by incorporating more structure and prior knowledge into their model architectures and loss functions. Differentiable Programming has emerged as a leading paradigm for systematically integrating converging domains of machine learning and scientific computing based on a shared infrastructure that is built on automatic differentiation of complex computer programs. This workshop will bring on board leading figures in differentiable programming for modeling and control of dynamical systems. Furthermore, the workshop will provide hands on coding examples of these emerging methods.

3. Computation for Real World Control Systems

Monday, October 2nd, 1:00 PM - 5:00 PM
Room: Cobalt

Organizer: Daniel Y. Abramovitch

ABSTRACT:
Computation is an essential component of implementing any real-world control system, but the details of how to make this work are often either left to the individual contributors to figure out or handed off to turn-key vendors. This workshop intends to provide insights, methods, and concrete examples into three major pieces of this subject. First, the workshop will present recent tutorial material (ACC 2023) from the author on real-time computing issues for control systems. This material explains the principal factors affecting the four computing chains inside a feedback system. After this overview, the workshop will spend time on an often-neglected area of computation for control system measurements, whether they be used in the control loop operation or in the system identification used in model building for control. Finally, the workshop will hone in on specific programming methods and components in the controller itself, describing efficient implementation methods and structures. Together these three thrusts should equip the participant with tools that they can apply almost immediately in their work.

4. Collaborative Robotics, Controls, and Machine Learning for Automated Visual Inspection of Complex Parts and Surfaces

Monday, October 2nd, 1:00 PM - 5:00 PM
Room: Lapis

Organizer: Colin Acton, SangYoon Back, Norawish Lohitnavy, Arun Nandagopal and Xu Chen

ABSTRACT:
This workshop focuses on automating surface inspection in manufacturing and developing a collaborative robotic manipulation intelligence. Subject matter experts from the industry (GE Research, GKN Aerospace, Gray Matter Robotics) will be invited to navigate the problem space and define key research problems. Then, we present successful experiences developing engineering solutions at the intersection of robotics, controls, data collection, and machine learning. The surface inspection system comprises a robotic manipulator (UR5e) with a custom end of arm tool consisting of a 61MP mirrorless camera with macro lens, an RGBD camera, and a 384-LED lighting array. We discuss the proposed system, results of experiments in defect classification on critical materials, and demonstrate the tools and novel control methods designed to perform subtasks involved in robotic surface inspection, including automated waypoint generation from a surface model of the part, intelligent path planning between imaging poses, and rejection of environmental disturbances such as illuminance variation across the part surface, motion-blur, and out-of-focus regions.

Tutorial Sessions, 2nd & 3rd October (Mon & Tue)

A Tutorial on State Space Models for Real-Time Control of Mechatronic Systems and a Discussion of Discretization Effects

Tuesday October 3rd, during PM Technical Session
Room: Lapis

ABSTRACT:
This tutorial session has two distinct, but equally important sections. The first talk will focus on the author’s state space structures, the Biquad State Space (BSS) and the Bilinear State Space (BLSS). These two structures have shown some remarkable advantages in the modeling and control of mechatronic systems, including numerical stability and model explainability. Furthermore, they have the remarkable property that the states of digital BSS and BLSS structures correspond to the states of the analog BSS and BLSS structures, at the outputs of each biquad or bilinear section. This gives the control engineer the ability to connect their digital model far more closely to the physical system, allowing digital “scope probes” to compare the same signals as one might get from the physical system. For the BSS and the BLSS to do their state-preserving magic, they must be discretized one block at a time. This flies in the current dogma that requires only a hold equivalent, typically a zero-order hold (ZOH) equivalent, can correctly discretize a feedback system. We will delve into the historical nature of this equivalent, as well as the conditions under which it represents the “exact” answer. We will also go through the handful of simple analytic examples used before one simply gives up and uses numerical means. We will also delve into what is lost in the name of mathematical exactitude. We finish with a comparison of the relative accuracy of discrete state-space forms for some illustrative examples, making the case that in many (and perhaps most) modern digital control systems, it may be worth giving up on mathematical exactitude in favor of having a very close system representation that can be understood and debugged.

Presentations:

1. PA Tutorial on the Biquad and Bilinear State Space Structures

ABSTRACT:
This tutorial paper will focus on the author’s state space structures, the Biquad State Space (BSS) and the Bilinear State Space (BLSS). These two structures have shown some remarkable advantages in the modeling and control of mechatronic systems, including numerical stability and model explainability. Furthermore, they have the remarkable property that the states of digital BSS and BLSS structures correspond to the states of the analog BSS and BLSS structures, at the outputs of each biquad or bilinear section. This gives the control engineer the ability to connect their digital model far more closely to the physical system, allowing digital “scope probes” to compare the same signals as one might get from the physical system.

2. A Discussion on Discretization and Practical Tradeoffs of the ZOH Equivalent

ABSTRACT:
This tutorial paper will focus on revisiting issues of discretization. The accepted, theoretically correct method of discretizing a linear, time-invariant model in a feedback loop is with a hold equivalent, most commonly the zero-order hold (ZOH) equivalent of the entire plant model. However, for the BSS and the BLSS to do their state-preserving magic, they must be discretized one block at a time. This flies in the current dogma that requires only a hold equivalent, typically a zero-order hold (ZOH) equivalent, can correctly discretize a feedback system. We will delve into the historical nature of this equivalent, as well as the conditions under which it represents the “exact” answer. We will also go through the handful of simple analytic examples used before one simply gives up and uses numerical means. We will also delve into what is lost in the name of mathematical exactitude. We finish with a comparison of the relative accuracy of discrete state-space forms for some illustrative examples, making the case that in many (and perhaps most) modern digital control systems, it may be worth giving up on mathematical exactitude in favor of having a very close system representation that can be understood and debugged.