Annual Review of Control, Robotics, and Autonomous Systems - Early Publication

This article reviews the literature on urban air mobility (UAM), examining both the research challenges it presents and the transformative opportunities that make these challenges worth addressing. While UAM has historical precedents, the current iteration is born of novel aircraft technology, primarily electric vertical takeoff and landing (eVTOL) and electric short takeoff and landing (eSTOL) aircraft. These advances raise new questions in aerodynamics, control, and integration with urban infrastructure. We explore several key research areas, including aircraft design, vertiport development, network planning, and air traffic management. We also address the scalability challenges in air traffic management for high-density UAM operations and the potential of autonomous and remotely piloted systems. If new aircraft are to birth a new urbanism, they will do so by integrating aircraft engineering and computational intelligence in control, systems, robotics, and human factors.

The development of software agents that can autonomously take actions to achieve goals has been a long-standing foundational objective in the field of AI. Recent advances in generative AI have given rise to a new class of agents. These advances have opened up the possibility of developing agents that can augment knowledge work in finance that primarily involves the cognitive processing of information by skilled humans. In this article, we bring these fields together. We break down the specific challenges in knowledge work in finance, review the current literature on generative AI agents, and identify potential directions for research and development that would help realize the potential of generative AI agents in finance. We conclude by proposing a framework to delineate the levels of autonomy for AI agents in the context of knowledge work, and an architecture for human–AI collaboration that can pave the path for progressively increasing the autonomy of agents.

This article is an attempt to reexamine the evolution of robotics research at the University of Pennsylvania and all that it entailed, covering the successes and struggles of PhD students, postdoctoral researchers, and a few dedicated faculty between 1972 and 2000. In 1945, Penn's Moore School of Electrical Engineering was famous for developing the first electronic digital computer, ENIAC, but after a few years, the research had diminished. In 1972, a new Department of Computer and Information Science was formed. I came to Penn from Stanford University's Artificial Intelligence Laboratory full of energy, enthusiasm, and the goal of establishing a similar lab at Penn. This article demonstrates the creativity and ingenuity of young professionals at the General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory. We built hardware. We built software. We collaborated with psychologists and electrical and mechanical engineers and tried to build a community of roboticists. Our curiosity led us to build new vision and tactile systems and to investigate cooperative robotics systems on the ground and in the air. We used computational models supported and verified by experiments. Today, I am proud to say that almost all of the students who cycled through the GRASP Lab are successful in academia or industry.

Linear integrators are well known for their ability to counter static forces and improve low-frequency disturbance rejection properties in control systems. However, linear integrators introduce phase lag, which is a frequency-dependent time shift or delay. Since the early introduction of the Clegg integrator, nonlinear integrators have held the promise of providing phase advantages over linear integrators when evaluated from the perspective of a describing function. This could potentially reduce delay and therefore provide the means to surpass linear design limitations; for example, overshoot and settling times can be reduced or even avoided. In addition, loop gains can be increased, which improves the low-frequency disturbance rejection properties. For five nonlinear integrators—the Clegg integrator, a generalized first-order reset integrator called the constant-in-gain–lead-in-phase (CgLp) element, a hybrid integrator–gain system, a variable gain integrator, and a split-path integrator—this article provides a comparison and overview of recent developments in stage motion control. Benchmark examples are taken from the industrial practice of wafer scanners, which form the pivotal machines used in the manufacturing of computer chips.

This article reviews two techniques that use delay for control: time-delay approaches to control problems (which initially may be free of delays) and the intentional insertion of delays into the feedback. We begin with a now widely used time-delay approach to sampled-data control. In networked control systems with communication constraints, this is the only method that accommodates transmission delays larger than the sampling intervals. We present a predictor-based design that enlarges the maximum allowable delay, which is important for practical implementations. We then discuss methods that use artificial delays via simple Lyapunov functionals that lead to feasible linear matrix inequalities for small delays and simple sampled-data implementations. Finally, we briefly present a new time-delay approach—this time to averaging. Unlike previous results, this approach provides the first quantitative bounds on the small parameter, making averaging-based control (including vibrational and extremum-seeking control) reliable.

The ability of a robot to build a persistent, accurate, and actionable model of its surroundings through sensor data in a timely manner is crucial for autonomous operation. While representing the world as a point cloud might be sufficient for localization, denser scene representations are required for obstacle avoidance. On the other hand, higher-level semantic information is often crucial for breaking down the necessary steps to autonomously complete a complex task, such as cooking. So the looming question is, What is a suitable scene representation for the robotic task at hand? This survey provides a comprehensive review of key approaches and frameworks driving progress in the field of robotic spatial perception, with a particular focus on the historical evolution and current trends in representation. By categorizing scene modeling techniques into three main types—metric, metric–semantic, and metric–semantic–topological—we discuss how spatial perception frameworks are transitioning from building purely geometric models of the world to more advanced data structures incorporating higher-level concepts, such as the notion of object instances and places. Special emphasis is placed on approaches for real-time simultaneous localization and mapping, their integration with deep learning for enhanced robustness and scene understanding, and their ability to handle scene dynamicity as some of the hottest topics of interest driving robotics research today. We conclude with a discussion of ongoing challenges and future research directions in the quest to develop robust and scalable spatial perception systems suitable for long-term autonomy.

The development of control methods based on data has seen a surge of interest in recent years. When applying data-driven controllers in real-world applications, providing theoretical guarantees for the closed-loop system is of crucial importance to ensure reliable operation. In this review, we provide an overview of data-driven model predictive control (MPC) methods for controlling unknown systems with guarantees on systems-theoretic properties such as stability, robustness, and constraint satisfaction. The considered approaches rely on the fundamental lemma from behavioral theory in order to predict input–output trajectories directly from data. We cover various setups, ranging from linear systems and noise-free data to more realistic formulations with noise and nonlinearities, and we provide an overview of different techniques to ensure guarantees for the closed-loop system. Moreover, we discuss avenues for future research that may further improve the theoretical understanding and practical applicability of data-driven MPC.