Andreea Bobu
I'm an Assistant Professor at MIT in AeroAstro and CSAIL. I lead the Collaborative Learning and Autonomy Research Lab (CLEAR Lab), where we develop autonomous agents that learn to do tasks for, with, and around people. Our goal is to ensure that these agents' behavior is consistent with human expectations, whether they interact with expert designers or novice users.
My work looks at: 1) getting the right data to supervise agents, whether directly from people or via priors; 2) enabling humans and robots to efficiently and interactively arrive at shared task representations; 3) quantifying and addressing misalignment caused by different human modeling choices. I ground my work in experiments and user studies with AI systems like assistive robot arms or LLMs, and draw upon methods from deep learning, mathematical human modeling, inverse reinforcement learning, and Bayesian inference.I obtained my Ph.D. in Electrical Engineering and Computer Science at UC Berkeley with Anca Dragan. You can read my thesis here. Before MIT, I was also a Research Scientist at the Robotics & AI Institute and an intern at NVIDIA in the Robotics Lab. Prior to my graduate degree, I received a B.S. in Computer Science at MIT.
news
Honored to receive theMIT Initiative for New Manufacturing Award with PI John Liu!
New preprint where wesurvey shared autonomy and discuss the role GenAI will play in the field.
Honored to receive theAmazon Research Award!
Two new arXiv papers on more efficient reward learning from multimodal human feedback -- one on learning fromlanguage and demonstrations in an offline regime, and one on learning online fromlanguage and corrections.
Our paper onGetting Aligned on Representational Alignment has been accepted to Transactions on Machine Learning Research (TMLR).
I had a wonderful conversation with Ella Lan from the AAAI Podcast Series about Human-Centric AI and Collaborative AI Systems. You can check out therecording on the AAAI podcast series Generations in Dialogue: Bridging Prspectives in AI.
At this year's RSS, I gave a talk on "Reading Between the Lines: Using LMs to Amplify Human Data in Robot Learning" at theWorkshop on Human-in-the-Loop Robot Learning. I also gave a talk on "Four Pillars of Human-Aligned Robot Representations" at theWorkshop on Learned Robot Representations.
Honored to receive theMIT Social and Ethical Responsibilities of Computing (SERC) Award!
Honored to receive theMIT Energy Initiative (MITEI) Award!
Honored to receive theMIT Generative AI Impact Consortium (MGAIC) Award!
Organizing a workshop where"GenAI meets HRI" atRSS 2025.
Read more about our lab vision for human-centric robot learning in thisFaculty Spotlight by MIT CSAIL.
Thank you MIT CSAIL for thisMeet the Mind profile on our lab!
Our paper on"Input-Adaptive Allocation of LM Computation" was accepted to ICLR 2025!
Organizing a workshop on"Long-term Human Motion Prediction" atICRA 2025.
Organizing a workshop on"Human-Centered Robot Learning in the Era of Big Data and Large Models" atICRA 2025.
I gave a talk at theWorkshop on Behavioral Machine Learning at NeurIPS.
I gave a keynote talk at theGlobal Summit on Open Problems for AI.
Check out this CSAIL Alliances podcast onRobot Conversations where I talk about how LLMs are changing the landscape of robotics research.
New preprint on"Goal Inference from Open-Ended Dialog"!
Our work on"Adaptive Language-Guided Abstraction from Contrastive Explanations" was accepted to CoRL 2024!
Organizing a workshop on"Robotic Tasks and How to Specify Them" atRSS 2024.
Organizing a workshop on"Mechanisms for Mapping Human Input to Robots" atRSS 2024.
I gave a talk at theCMU RI RoboLaunch Speaker Series, an outreach program for promoting robotics and AI research and education.
I gave an invited talk atETH.
Organizing a workshop on"Long-term Human Motion Prediction" atICRA 2024.
I gave a guest lecture in MIT's course onRobotics: Science and Systems, a course that I took as an undergraduate student 9 years ago.
Check out myTEDxMIT talk onWhy Robots Aren’t Superhuman in Our Human World!
I gave a guest lecture in CMU's course onInteractive Robotics.
Organizing a workshop on"Lifelong Learning and Personalization in Long-Term Human-Robot Interaction (LEAP-HRI)" atHRI 2024.
Our work on "Preference-Conditioned Language-Guided Abstraction" was accepted to HRI 2024!
Our work on "Aligning Human and Robot Representations" was accepted to HRI 2024!
Check out our new preprint where we discuss "Getting Aligned on Representational Alignment"!
I was awarded the Emerging Research Award at theInternational Symposium on the Mathematics of Neuroscience!
Organizing a workshop on"Interactive Learning with Implicit Human Feedback" atICML 2023.
New paper accepted to CDC on"Diagnosing and Augmenting Feature Representations in Correctional Inverse Reinforcement Learning", where we propose a method for repairing feature representations to better generalize across states!
I accepted a visiting Research Scientist position at theBoston Dynamics AI Institute starting August 2023!
I gave an invited talk at theCHAI Workshop.
I gave an invited talk at theStanford Robotics Seminar. Check out the talkhere!
I accepted an offer to join theDepartment of Aeronautics and Astronautics in theSchool of Engineering atMassachusetts Institute of Technology (MIT) starting Fall 2024!
I published a patent with the NVIDIA Robotics team on a new concept training technique for machine learning based on our paper on"Learning Perceptual Concepts by Bootstrapping from Human Queries"!
New paper accepted to ICML,"Diagnosis, Feedback, Adaptation: A Human-in-the-Loop Framework for Test-time Policy Adaptation", where we propose a human-in-the-loop framework for robots to recover from distribution shift failures.
Giving an invited talk at the Microsoft Research Seminar Series.
Organizing a workshop on"Aligning Robot Representations with Humans " at theConference on Robot Learning 2022.
Giving an invited talk at theRobotics Colloquium at UW.
Giving an invited talk at theNew Trends in Aerospace Seminar Series at MIT.
New paper accepted to HRI,"SIRL: Similarity-based Implicit Representation Learning", where we enable robots to learn salient feature representations by asking humans to gauge how similar different behaviors are.
I gave an invited talk at theRobotics Seminar at Cornell.
I was invited to speak at University of Utah'sCS 6960: Human-AI Alignment graduate course.
I gave an invited talk in theRobot Autonomy and Interactive Learning (RAIL) Lab at Georgia Tech.
Our paper onTime-Efficient Reward Learning via Visually Assisted Cluster Ranking has been accepted to theHuman-in-the-loop Learning (HILL) Workshop at NeurIPS 2022.
I gave an invited talk at theIllinois Robotics Seminar at UIUC.
I gave an invited talk in the Intelligent and Interactive Autonomous Systems Group (ILIAD) Group at Stanford.
I am grateful to have been selected for theRising Star in EECS Academic Career Workshop!
I gave an invited talk at the Workshop on Complex Feedback in Online Learning at ICML.
Co-organized a workshop on "Social Intelligence in Humans and Robots" at Robotics: Science and Systems 2022.
New paper accepted to IROS, "Teaching Robots to Span the Space of Functional Expressive Motion", where we enable people to efficiently teach robots expressive motions during task execution.
New paper accepted to Robotics and Automation Letters, "Learning Perceptual Concepts by Bootstrapping from Human Queries". In this work, we ask for human input to learn a low-dimensional variant of the perceptual concept, then use it to generate a larger data set for learning the concept in the high-dimensional space.
I gave an invited talk at Apple's AI/ML seminar.
I was named aRobotics: Science and Systems Pioneer!
Our paper on Aligning Robot Representations with Humans has been accepted to the Workshop on Collaborative Robots and the Work of the Future, at ICRA 2022.
Our paper"Inducing Structure in Reward Learning via Feature Learning" was accepted at the International Journal of Robotics Research.
I gave an invited talk atAware-learning: how to benefit from priors, a workshop at theConference on Decision and Control.
I gave an invited talk at MIT's Interactive Robotics Group.
I was named anApple Scholar in AI and Machine Learning (AI/ML)!
Our journal paper "Quantifying Hypothesis Space Misspecification in Learning from Human-Robot Demonstrations and Physical Corrections" received an Honorable Mention for the 2020 IEEE Transactions on Robotics King-Sun Fu Memorial Best Paper Award.
Our paper on reward learning using feature traces, a novel type of human input, was nominated for a Best Paper Award at the ACM/IEEE International Conference on Human-Robot Interaction 2021.
Our paper "Situational Confidence Assistance for Lifelong Shared Autonomy" was accepted at the IEEE International Conference on Robotics and Automation 2021.