Postdoctoral Associate (Reinforcement Learning for Embodied Intelligence and Human Learning)

Project Overview

We are seeking one Postdoctoral Associate in Computer Science to join the research program "Mens, Manus and Machina—How AI Empowers People, Institutions and Cities in Singapore (M3S)." Successful applicants for this position will have the opportunity to work on cutting-edge research at the intersection of reinforcement learning (RL), world models, embodied intelligence, and human learning. The research targets a new generation of intelligent agents that learn through sequential interaction, and is organized around two pillars united by a shared reinforcement learning foundation.

Embodied intelligence. This part focuses on robotic agents operating in unstructured 3D environments (homes, offices, industrial sites, shopping malls). The work centres on (a) model-based and hybrid RL algorithms built on multi-modal world models (vision, language, proprioception, action) that support long-horizon planning, principled exploration, and sim-to-real transfer; (b) compositional and hierarchical world models that translate higher-order, qualitative human goals into grounded sequences of manipulation and navigation actions, with suitable adaptation to account for task-specific performance constraints (e.g., accuracy, completion deadline); and (c) learning from human feedback, e.g., imitation, inverse RL, RLHF, and interactive/intervention-based learning, so that robot behaviour remains aligned with human intent under partial observability.

Human learning. This part develops an adaptive learning system in which an AI tutor estimates each student's evolving knowledge state from their interaction history (responses, errors, time-on-task) and personalizes what to deliver next, e.g., lessons, exercises, hints, and review items, to maximize long-term learning gains. RL is the natural framing here: the tutor is a sequential decision-maker, the learner's knowledge state is the (partially observed) environment, and the world model is a learner model that predicts how a student's mastery will evolve under different pedagogical actions. Notably, the tutor’s actions can encompass not just the temporal phasing of knowledge content (the curriculum) but also the 3D presentation of such content (including AI-based generation of 2D/3D vision-language embodiments of such content).  The research advances knowledge tracing under partial observability, curriculum and content sequencing, and learning from human feedback (instructor preferences, learner self-reports) to keep recommendations pedagogically aligned.

This program is led by distinguished scholars, including Profs. Sanjay Sarma, Daniela Rus and Jinhua Zhao from MIT; and Prof. Archan Misra from Singapore Management University.

Responsibilities

• Conduct and coordinate research on new RL algorithms, world models, and human-in-the-loop learning methods, applied to both embodied agents (robots in 3D environments) and adaptive learning systems (AI tutors that model and guide human learners).
• Develop learner models and content-recommendation policies that estimate a student's evolving knowledge state and sequence lessons, exercises, and hints to maximize long-term learning outcomes, with suitable adaptations for specific student segments (child vs. adult learners)
• Publish high-quality research results in top-tier venues spanning machine learning, reinforcement learning, embodied AI, and AI in education (e.g., NeurIPS, ICML, ICLR, ACL, CoRL, IEEE ICRA/IROS, AIED, EAAAI).
• Participate in the mentorship and training of graduate and undergraduate students at MIT and in Singapore, and research engineers based in Singapore.
• Participate in, and lead, small technical teams in creating research prototypes and technology demonstrators for external research partners
• Assist in grant writing, project management, and other administrative duties related to research activities.

Requirements

• Ph.D. in Computer Science or a closely related field, with research focus in Reinforcement Learning, Machine Learning, Embodied AI, or AI for Education.
• Strong background in deep reinforcement learning, with hands-on experience training agents in simulated, physical, or human-interactive environments.
• Experience with world models or learned predictive models, covering physical dynamics (e.g., latent dynamics for robotics) or learner dynamics (e.g., knowledge tracing, cognitive/student models).
• Strong publication record in top-tier ML/RL, embodied AI, or AI-in-education venues (e.g., NeurIPS, ICML, ICLR, ACL, CoRL, ICRA/IROS, AIED).
• DESIRABLE: Experience deploying learning systems with real users, sim-to-real for robots, classroom/online-platform deployment of adaptive tools, or efficient on-device inference.
• Excellent communication and collaboration skills.

To find out more about this role, please contact Professor Archan Misra (archanm@smu.edu.sg) and Dr Alok Prakash (alok.prakash@smart.mit.edu).

To apply, please visit our website at: https://portal.smart.mit.edu/careers/career-opportunities

Interested applicants are invited to send in their full CV/resume, cover letter and list of three references (to include reference names and contact information). We regret that only shortlisted candidates will be notified.