Keynote Lectures

Abstract
Constraints refer to limits imposed on state and control variables which must be satisfied during system operation. Examples of constraints include but are not limited to range and rate actuator limits, pressure and temperature safety limits and obstacle avoidance requirements. With the continuing trends towards growing system autonomy, improved performance and downsizing, constraint handling and limit protection functions are becoming increasingly important to enable engineered systems to operate safely at the “limits”.
The presentation will introduce the reference governor, an add-on predictive safety supervision algorithm that monitors and modifies, if it becomes necessary, commands passed to the nominal system to ensure that constraints are satisfied. Approaches to the design and implementation of reference governors will be described along with the supporting theory. The potential for the applications of reference governors to automotive and aerospace systems will be illustrated with several examples.
Recent extensions of reference governors which include controller state and reference governors, action governors, and feasibility governors which supervise Model Predictive Controllers will be described. The learning reference governor, which integrates learning into the reference governor operation, to handle constraints in uncertain systems will also be introduced.
The presentation will end with speaker’s perspectives on challenges and opportunities in handling constraints in increasingly autonomous systems.

Abstract
In the past 10 years or so, machine learning has made great progress, in terms of both theories and applications. Among various machine learning methods, deep learning is one of the most widely used methods. Significant advance has been made in speech recognition, object recognition and detection as well as in many other applications using deep learning. In this talk, we present a learning algorithm that is mathematically equivalent to the well-known back-propagation learning algorithm in neural networks but have the following advantages over the back-propagation algorithm. (1) It does not require a feedback network to back-propagate the errors. This makes its implementations, especially implementations on silicon, much simpler. (2) It is biologically plausible as all information needed for synapses to adapt is available in a biological neuron. Hence, artificial neural networks indeed mimic biological neural networks. (3) It can be applied to general dynamic systems, that is, many dynamic systems can adapt in a way similar to neural networks. This property allows us to introduce “intelligence” into a large class of engineering systems. In particular, we consider adaptation in control systems using the proposed learning algorithm, including adaptive state feedback, adaptive PID controller, and model reference adaptive control.

Abstract
Machine Learning and Optimization techniques are now widely used in many scientific disciplines. Using Machine Learning and Optimization in Robotics is a very challenging task as robots may be expensive and fragile, and the time and effort to collect data is, in general, high. Considering these premises, we have developed several techniques that through the use of simulators, and adapted learning/optimization algorithms, that use the data very efficiently, make the use of Learning and Optimization in Robotics an effective option for hand-coded approaches. These techniques have been applied to solve perception, reasoning and behavior level and multi-agent robotic tasks.
This talk will present some of these techniques, in different contexts, namely those related to ML/optimization based development of robotic skills, data preparation, Q-Batch update rule, multi-agent learning, adapted interfaces and mixed deep learning/heuristic classification.

Abstract
We explore the potential of using the internal model principle of control theory to explain the cerebellum, a major component of the brain. The cerebellum is involved in motor control, motor learning, posture and balance, gait control, eye movement, language regulation, emotion regulation, etc. It has been described as regulating and coordinating all movements of precision. Neuroscientists hypothesize that the cerebellum contains forward or inverse models of the systems it regulates. Unfortunately, conclusive experimental proof that this hypothesis is correct has not been forthcoming, despite a 30 year effort. This suggests that the problem is particularly challenging. But it also suggests that perhaps it is time for the hypothesis to be re-examined. In this talk we describe an alternative approach to modeling the cerebellum, based on a hypothesis that its primary function is disturbance rejection. We present an overview of the key methods in control theory that allow us to model the cerebellum, some of which have only been developed in the last five to ten years. Evidence of the efficacy of our approach is given in terms of the slow eye movement systems, the optokinetic system, the saccadic system, and visuomotor adaptation. We conclude the talk with implications for robotics.