PLAN-FM @ AAAI 2026

Schedule

9:00 AM to 5:00 PM (Singapore Time)
Location: Topaz 220 – 225 (map)

Time Program

9:00Opening Remarks (slides)
9:10Tutorial by Shirin Sohrabi
A Brief Introduction to AI Planning (slides)
9:50Tutorial by Wenjun Li
Real-World Applications and Benchmarks (slides)
10:30Coffee Break
11:00 Invited Talk by David Hsu

Towards Open-World Robot Planning: Rich Priors and Failure Resilience

A hallmark of intelligence is the ability to do the right thing in myriad unfamiliar situations. The classic model-based approach to robotics draws a sharp boundary between the closed, known world and the open, unknown world: robot performance is guaranteed only in known situations. Data-driven robot foundation models, with their vast common-sense knowledge, have blurred this boundary and dramatically expanded robot capabilities in the open world. In this talk, I will argue that successful robot planning in the open world hinges on two core capabilities: access to rich, diverse information and resilience under unexpected failures. To illustrate these ideas, I will present our recent work: (i) robot navigation using floor plans, hand-drawn maps, ... (ii) robot operating home appliances by reading manuals, and (iii) detect and recover from unexpected events.
11:45Contributed Talk:
SAMKE: An Open-Ended Autonomous Foundation-Model-Based Agent for Meta-Knowledge Discovery
12:00Contributed Talk:
Next-Latent Prediction Transformers Learn Compact World Models
12:15Contributed Talk:
Rethinking Reward Models! A Conceptual Framework for Enhancing LLM Reasoning through Intrinsic Traits
12:30Lunch
14:00 Invited Talk by Pulkit Verma

Teaching LLMs to Plan: From Chain-of-Thought Instruction Tuning to Collaborative Constraint Translation

Large Language Models fail at planning while automated planners remain inaccessible to domain experts. This talk presents two complementary neuro-symbolic solutions. I show how logical chain-of-thought instruction tuning with symbolic verifier feedback improves LLM planning by modifying model parameters (not just prompts) using detailed feedback to teach logical reasoning. Llama-3 improves from 28% to 94% plan validity, with gains across multiple models and domains. I then present a collaborative framework that translates natural language guidance into formal planning constraints, enabling domain experts to guide planning without PDDL knowledge. The system achieves 90% translation accuracy and enables 40-50% cost reductions on logistics problems, though effectiveness varies by domain. Improved planning reasoning directly enables better translation of human intent. Together, these demonstrate how to make planning both reliable and accessible through neuro-symbolic collaboration. I discuss results, limitations, and future directions.
14:30Contributed Talk: Remote
Metrics for Holistic Evaluation of LLM Reasoning about Action, Change, and Planning
14:45Contributed Talk: Remote
ProofNet++: A Neuro-Symbolic System for Formal Proof Verification with Self-Correction
15:00 Invited Talk by Asim Munawar

Reasoning with LLMs: Neuro-Symbolic AI & Agents to Rescue?

In this talk, I will analyze the reasoning limits of LLMs and discuss how neuro-symbolic techniques and agentic workflows can bridge the gap between fluent language generation and reliable decision-making.
15:30Coffee Break
16:00
Closing Remarks
Poster Session
- Planning Beyond Perception: Benchmarking LLM- and VLM-Based Reasoning for Autonomous Driving
- Rethinking Reward Models! A Conceptual Framework for Enhancing LLM Reasoning through Intrinsic Traits
- Next-Latent Prediction Transformers Learn Compact World Models
- SAMKE: An Open-Ended Autonomous Foundation-Model-Based Agent for Meta-Knowledge Discovery and Learning

Invited Speakers

David Hsu
National University of Singapore

Bio:
David Hsu is a professor of computer science and the Director of Smart Systems Institute at the National University of Singapore (NUS). He is an IEEE Fellow. His research lies in the intersection of robotics and AI. In recent years, he has been working on robot planning and learning under uncertainty for human-centered robots. His work won multiple international awards, including, most recently, Test of Time Award at Robotics: Science & Systems (RSS) in 2021 and IJCAI-JAIR Best Paper Prize in 2022. He has chaired or co-chaired several international robotics conferences, including WAFR 2010, RSS 2015, ICRA 2016, and CoRL 2021. He served on the editorial boards of Journal of Artificial Intelligence Research and International Journal of Robotics Research. He is currently an Editor of IEEE Transactions on Robotics.

Pulkit Verma
Indian Institute of Technology Madras

Bio:
Pulkit Verma is an Assistant Professor at the Department of Computer Science and Engineering, IIT Madras. Prior to this, he was a Postdoctoral Associate in the Interactive Robotics Group at the Massachusetts Institute of Technology, where he worked with Prof. Julie Shah. His research focuses on the safe and reliable behavior of taskable AI agents. He investigates the minimal set of requirements in an AI system that would enable a user to assess and understand the limits of its safe operability. He received his Ph.D. in Computer Science from the School of Computing and Augmented Intelligence, Arizona State University, where he worked with Prof. Siddharth Srivastava. Before that, he completed his M.Tech. in Computer Science and Engineering at Indian Institute of Technology Guwahati with Prof. Pradip K. Das. He was awarded the ICAPS 2025 Outstanding Dissertation Award at ICAPS 2025, AAAI/ACM SIGAI Innovative AI Education Award at AAAI 2025's EAAI Symposium, Graduate College Completion Fellowship at ASU in 2023, Post Graduation Scholarship from the Government of India in 2013 and 2014, and received the Best Demo Award at the International Conference on Autonomous Agents and Multiagent Systems (AAMAS) in 2022.

Asim Munawar
IBM Research

Bio:
Dr. Asim Munawar is a Project Lead at IBM’s Watson Research Center in New York, where he heads efforts to enhance reasoning, planning, and agentic workflows in enterprise-scale large language models. With over 15 years of experience in AI—more than a decade of it at IBM Research—he has held key leadership roles, including Manager and Program Director for Neuro-Symbolic AI. Dr. Munawar earned his Ph.D. from Hokkaido University, Japan, and has authored over 80 peer-reviewed publications. He is an inventor on 20+ U.S. patents and a frequent keynote and invited speaker at top venues such as IJCAI, ICSE, and ACMSE. He also serves on advisory boards for the National Center of Artificial Intelligence in Pakistan and the Centaur AI Institute in the U.S.

Accepted Papers

ProofNet++: A Neuro-Symbolic System for Formal Proof Verification with Self-Correction,
Murari Ambati
Planning Beyond Perception: Benchmarking LLM- and VLM-Based Reasoning for Autonomous Driving, Keonvin Park,
Seunghyeok Hong
Rethinking Reward Models! A Conceptual Framework for Enhancing LLM Reasoning through Intrinsic Traits,
Het Riteshkumar Shah, Megha Sundriyal
Next-Latent Prediction Transformers Learn Compact World Models,
Jayden Teoh, Manan Tomar, Kwangjun Ahn, Edward S. Hu, Pratyusha Sharma, Riashat Islam, Alex Lamb, John Langford
Metrics for Holistic Evaluation of LLM Reasoning about Action, Change, and Planning,
Anil B Murthy, Jaron Mink, Lindsay Sanneman
SAMKE: An Open-Ended Autonomous Foundation-Model-Based Agent for Meta-Knowledge Discovery and Learning,
Nitin Vetcha, Dianbo Liu

Call for Papers

The 2nd PLAN-FM Bridge Program invites cutting-edge research on leveraging Foundational Models (e.g. large language models, large reasoning models, multi-modal models) for multi-step reasoning and planning. Since 2025, large reasoning models and RL-tuned variants have advanced markedly; hybrid pipelines (LLM + classical planner/verifier) are maturing; and new cross-domain benchmarks expose persistent gaps in executability and robustness. We solicit papers on scalable, grounded, and verifiable planning with foundational models across the topics below.

Topics of interest

FM for Planning & Decision-Making: Prompting/training for long-horizon control, plan decomposition, and search efficiency.
Planning for Agent Orchestration: Efficient, accurate, and trustworthy planning solutions in agentic frameworks and applications.
Embodied & Multi-Agent Planning: Robotics, autonomy, coordination, and human-in-the-loop planning under real-world constraints.
Reliability, Safety & Guarantees: Verified/executable plans, constraint satisfaction, formal checks, and failure analysis.
Planner-in-the-Loop Tool Use: Integration with symbolic planners, search (A*, MCTS), simulators, program synthesis, knowledge bases, and environment feedback.
Benchmarks & Shared Resources: Cross-domain datasets/simulators; standardized tasks for reproducible, apples-to-apples comparisons.
Plan Quality & Stress-Testing: Metrics and toolkits for executability, optimality, generalization, and efficiency; leaderboards and robustness suites.
Next-Gen FM Ingredients: Memory/state tracking, long-context handling, world models, multimodal grounding, and modular architectures for planning.

Submission instructions

Papers should be formatted according to the AAAI-26 two-column format (author kit). Submissions are handled on OpenReview. We welcome several submission types:

Papers – Short (up to 4) and long (up to 8 pages, excluding references) papers, describing novel ideas, perspectives, or early research results.
Extended Abstracts – up to 2 pages (excluding references), summarizing late-breaking results, preliminary findings, or challenges to provoke discussion.
System Demonstrations – up to 4 pages (plus references), showcasing innovative systems or prototypes (include a description and optionally a screenshot or link).