Leslie Pack Kaelbling is an American roboticist and the Panasonic Professor of Computer Science and Engineering at the Massachusetts Institute of Technology. She is widely recognized for adapting partially observable Markov decision processes from operations research for application in artificial intelligence and robotics. Her work has shaped how autonomous systems reason about uncertainty and act in dynamic environments.
Kaelbling received the IJCAI Computers and Thought Award in 1997 for applying reinforcement learning to embedded control systems and developing programming tools for robot navigation. In 2000, she was elected as a Fellow of the Association for the Advancement of Artificial Intelligence. Her research spans decision-making under uncertainty, machine learning, and sensing, with a focus on enabling robots to operate reliably in the real world.
Early Life and Education
Kaelbling earned an A.B. in Philosophy in 1983 from Stanford University, followed by a Ph.D. in Computer Science in 1990, also from Stanford. During her graduate studies, she was affiliated with the Center for the Study of Language and Information, where she explored connections between logic, reasoning, and action. Her doctoral work laid the groundwork for her later contributions to probabilistic planning and control.
Her interdisciplinary background - combining philosophy with computer science - informed her approach to AI, particularly in framing intelligent behavior as a problem of acting under uncertainty. This perspective became a hallmark of her career, influencing both theoretical frameworks and practical robotic systems.
Career Trajectory
After completing her Ph.D., Kaelbling worked at SRI International and the affiliated robotics spin-off Teleos Research. At SRI, she contributed to early efforts in robot navigation and embedded control, developing tools that allowed robots to make decisions based on noisy sensor data. Her time at Teleos Research bridged academic research and commercial applications, giving her insight into the practical challenges of deploying AI systems.
In the mid-1990s, Kaelbling joined the faculty at Brown University, where she established a research group focused on reinforcement learning and probabilistic robotics. At Brown, she collaborated with colleagues on foundational papers that defined how robots could learn from interaction and plan under partial observability. She remained at Brown until 1999, when she moved to the Massachusetts Institute of Technology.
At MIT, Kaelbling became the Panasonic Professor of Computer Science and Engineering, a position she continues to hold. Her lab at MIT's Computer Science and Artificial Intelligence Laboratory has produced influential work on task and motion planning, hierarchical decision-making, and integrated perception-action systems. She has mentored numerous students who have gone on to prominent careers in academia and industry.
Contributions to Reinforcement Learning
Kaelbling's early work in reinforcement learning helped establish the field as a core area of AI research. Her 1996 survey, "Reinforcement Learning: A Survey," co-authored with Michael L. Littman and Andrew W. Moore, remains one of the most cited references in the field. The survey systematically organized existing algorithms, theoretical results, and applications, providing a roadmap for subsequent research.
Her 1997 IJCAI Computers and Thought Award recognized her contributions to applying reinforcement learning to embedded control systems. This work demonstrated that robots could learn policies directly from experience, without requiring explicit models of their environments. She also developed programming tools for robot navigation that allowed researchers to specify high-level goals while the system handled low-level uncertainty.
Kaelbling's approach emphasized the importance of representing uncertainty explicitly. Rather than assuming perfect knowledge, her systems maintained probability distributions over possible states and used these to guide action selection. This perspective was particularly influential in robotics, where sensors are noisy and actuators are imprecise.
Partially Observable Markov Decision Processes
A central contribution of Kaelbling's career is the adaptation of partially observable Markov decision processes (POMDPs) from operations research to AI and robotics. POMDPs provide a mathematical framework for decision-making when an agent cannot directly observe the full state of its environment. Kaelbling and her collaborators showed how these models could be used for robot navigation and planning.
Her 1998 paper, "Planning and acting in partially observable stochastic domains," co-authored with Littman and Anthony R. Cassandra, introduced efficient algorithms for solving POMDPs and demonstrated their application to mobile robot navigation. This work bridged theoretical AI and practical robotics, showing that POMDPs could handle the uncertainty inherent in real-world sensing.
Kaelbling's related work on discrete Bayesian models for mobile-robot navigation, with Cassandra and James A. Kurien, provided concrete methods for updating beliefs about a robot's location using sensor data. These techniques became foundational for later developments in simultaneous localization and mapping (SLAM) and probabilistic robotics more broadly.
Journal of Machine Learning Research
In the spring of 2000, Kaelbling played a pivotal role in the open access movement in AI. She and two-thirds of the editorial board of the Kluwer-owned journal Machine Learning resigned in protest over its pay-to-access archives and limited financial compensation for authors. The resignation was a coordinated action to challenge the commercial publishing model that restricted access to research.
Kaelbling co-founded the Journal of Machine Learning Research (JMLR) and served as its first editor-in-chief. JMLR was a peer-reviewed open access journal that allowed researchers to publish articles for free, retain copyright, and make archives freely available online. This model was revolutionary at the time and has since become a standard for many AI publications.
In response to the mass resignation, Kluwer changed its publishing policy to allow authors to self-archive their papers online after peer review. Kaelbling noted that this policy was reasonable and would have made the creation of an alternative journal unnecessary. However, she observed that the policy change came only after the threat of resignations and the actual founding of JMLR, underscoring the importance of collective action in scholarly publishing.
Selected Works and Impact
Kaelbling's publication record includes several highly influential papers. Her 1996 survey on reinforcement learning, with Littman and Moore, appeared in the Journal of Artificial Intelligence Research and has been cited thousands of times. The 1998 POMDP paper with Littman and Cassandra, published in Artificial Intelligence, similarly became a cornerstone reference.
Her work on hierarchical task and motion planning, with Tomás Lozano-Pérez, addressed the challenge of combining discrete task-level planning with continuous motion planning. Their 2011 paper, "Hierarchical task and motion planning in the now," presented methods for interleaving planning and execution in real time, enabling robots to handle complex manipulation tasks. This line of research has influenced modern approaches to robot manipulation in both academia and industry.
Kaelbling also contributed to practical reinforcement learning in continuous spaces, with William D. Smart, in a 2000 paper presented at the International Conference on Machine Learning. This work addressed the challenge of scaling reinforcement learning to high-dimensional, continuous state spaces, a problem that remains central to modern deep reinforcement learning.
Legacy and Recognition
Kaelbling's influence extends beyond her own research. As a professor at MIT, she has shaped the next generation of roboticists and AI researchers. Her emphasis on principled probabilistic methods has become standard practice in robotics, influencing fields from autonomous driving to household robots.
Her election as a Fellow of the Association for the Advancement of Artificial Intelligence in 2000 recognized her sustained contributions to the field. The IJCAI Computers and Thought Award, one of the highest honors in AI, acknowledged her early career achievements. Her role in founding JMLR also had a lasting impact on how AI research is disseminated, promoting open access before it became a widespread movement.
Kaelbling's work continues to be relevant as AI systems increasingly operate in uncertain, real-world environments. Her foundational contributions to POMDPs and reinforcement learning underpin many modern approaches in robotics and decision-making, from warehouse automation to autonomous vehicles. Her career exemplifies the integration of theoretical rigor with practical application, a model that continues to inspire researchers in AI and robotics.