Safe Physical AI - IJCAI/ECAI 2026 Workshop

About the Workshop

Recent progress in deep learning and robot learning has brought the vision of autonomous, embodied systems operating in real-world environments significantly closer to reality. Robots and other physically situated AI systems are now demonstrating strong generalization, contact-rich manipulation and long-horizon reasoning capabilities. As these increasingly operate in human environments, ensuring both physical and cognitive safety becomes an urgent research priority.

This workshop focuses on Safe Physical AI: the study and development of intelligent systems that can act safely and reliably in the real world. We seek to identify fundamental challenges in verification, interpretability, uncertainty quantification, human–robot interaction, and alignment for physical AI systems. The workshop will unite researchers from robotics, deep learning, AI safety, planning, and AI governance to develop a shared conceptual and methodological foundation for safe embodied intelligence. By fostering interdisciplinary dialogue and mapping concrete research directions, the workshop aims to advance the scientific and ethical foundations required for trustworthy physical AI.

Key Objectives

To foster cross-disciplinary exchange between researchers in robotics, AI safety, machine learning, planning, control, ethics, and AI governance, encouraging collaboration and shared methodologies for safety assurance in embodied systems.
To define challenges and opportunities in developing safe, reliable, and interpretable embodied agents by analysing concrete safety and alignment issues in real-world interaction, manipulation, and long-horizon autonomy.
To develop a shared understanding of fundamental concepts, risk taxonomies, and evaluation strategies for assessing the safety and trustworthiness of embodied AI.
To investigate how AI methods—including reinforcement learning, uncertainty-aware models, causal reasoning, and formal verification—can contribute to safe decision-making and control under uncertainty.
To identify gaps in current architectures and design methodologies for engineering embodied agents that meet rigorous safety, robustness, and ethical requirements.

Call for Papers

This workshop invites submissions that develop methods, theory, empirical insights, benchmarks, and conceptual frameworks for Safe Physical AI: real-world intelligent systems that act safely and reliably under uncertainty. We welcome technical, empirical, theoretical, and position papers.

We welcome submissions of work that has been previously published or is under review at other venues. Authors should indicate this at submission time. Such submissions will be evaluated based on their relevance to the workshop themes.

Submission Formats

We accept short papers (4 pages excl. references) as well as extended abstracts (1 page excl. references). All submissions must use the two-column IEEE template. Authors of four-page papers will present their work in a seven-minute talk, with an additional seven minutes for questions and discussion. Authors of extended abstracts will present their work in two-minute lightning talks. Authors of short papers and extended abstracts are strongly encouraged to submit a poster and participate in the mentoring program. We strongly encourage interdisciplinary submissions, early-stage research papers, and contributions from graduate students.

Important Dates

Submissions open	March 26, 2026
Submission deadline	May 7, 2026, 23:59 CET
Author notification	June 7, 2026

Submit via OpenReview: openreview.net/group?id=ijcai.org/IJCAI-ECAI/2026/Workshop/SPAI

Relevant Topics

Safe decision making and action selection for embodied agents
Uncertainty quantification, risk estimation, and probabilistic verification
Interpretability of policies and world models, including mechanistic and behavioral approaches
Task specification, alignment, and intent understanding for physical agents
Monitoring and evaluation of untrusted models or untrusted hardware
Runtime verification, safety shields, and control-theoretic safety mechanisms
Bridging classical robot safety and statistical learning: safety beyond toy domains
Safe operation in social, contextual, and multi-agent environments
Stress testing, benchmarking, and evaluation methodologies
Contributions from ethics, law, and philosophy to physical AI safety
Interdisciplinary perspectives on risk taxonomies, safety vocabularies, and conceptual foundations
Hardware and computational design choices for safety in physical AI
Safe autonomy in real-world cyberphysical systems, including aerial, mobile, and manipulation platforms

Invited Speakers

George J Pappas

University of Pennsylvania

Machine Learning for Safe and Secure Cyber-Physical Systems

Andrea Bajcsy

Carnegie Mellon University

Interactive and Trustworthy Robotics

Sao Mai Nguyen

IP Paris

Robot Learning

Saskia Nagel

RWTH Aachen

AI Ethics, Philosophy of Technology

Interactive Formats

The workshop prioritizes active participation and knowledge exchange through multiple interactive formats designed to engage attendees across career stages and disciplines.

Poster Session

Poster sessions provide dedicated venues for early-career researchers to present their work and receive feedback from the broader community. Authors will submit one-page extended abstracts describing their research, with each presenter delivering a two-minute lightning talk to preview key ideas and encourage focused discussions at their poster. Authors of four-page short papers will also be given the opportunity to present a poster. Poster sessions are strategically scheduled during coffee breaks and between invited talks and panel discussions to maximize attendance and foster informal interactions.

Mentoring Program

To support early-career researchers and enhance inclusivity, we implement a structured mentoring program that pairs poster presenters with invited speakers or senior researchers based on shared interests. Authors can opt into the program when submitting extended abstracts or short papers. Mentors provide feedback on poster drafts and prioritize visiting their assigned mentees' posters during the session. Dedicated mentoring sessions during breaks facilitate discussions about research challenges, career development, and potential collaborations.

Panel Discussion

The panel discussion will convene invited speakers for a moderated conversation examining challenges and advancements in safe physical AI from multiple disciplinary perspectives. Attendees can submit questions through an online platform where others can vote to prioritize the most pressing topics.

Program Committee

Prof. Michael Fisher (University of Manchester) — Verification of Autonomous Systems; Trustworthy AI
Prof. Michael Beetz (University of Bremen) — Knowledge Representation and Reasoning; Robot Cognitive Architectures
Prof. Masoumeh Mansouri (University of Birmingham) — Hybrid Robot Intelligence; Social Robotics
Dr. Michaela Kümpel (University of Bremen) — Knowledge Systems and Human-Robot Interaction
Dr. Maciej Zajac (Polish Academy of Sciences) — Ethics of Autonomous Weapons Systems
Prof. Rania Rayyes (Karlsruhe Institute of Technology) — Developmental Robots, Dexterous Robots, Industrial Robotics
Prof. Darko Katic (Stuttgart Technical University of Applied Sciences) — Knowledge Representation and Reasoning; Surgical Robotics
Dr. Justin Shenk (Redwood Research) — Evaluations and risk assessments for frontier AI models
Prof. Barbara Hammer (University of Bielefeld) — Robust, Data-parsimonious, Fair Machine Learning
Prof. Daniel Neider (TU Dortmund) — Safety Verification of AI systems
Prof. Vera Schmitt (TU Berlin) — Trustworthy and Responsible NLP
Prof. Julius Schöning (Osnabrück UAS) — Intersections to Legal Issues
Dr. Tim Schrills (University of Lübeck) — Psychology, AI Act
Prof. Peter Fettke (DFKI) — Human-Centric AI, Economics
Prof. Hanna Drimalla (University of Bielefeld) — Multi-Modal Sensors, Human-Centric AI
Max Kroker (Independent Lawyer) — Law of AI
Dr. Daniel Winteler (Fraunhofer Society) — Law of AI

Organizers

Benjamin Alt

University of Bremen, Germany / Robotics Institute Germany (RIG)

Technical Director at the AICOR Institute for Artificial Intelligence and co-founder of AICOR Solutions, a startup for safe robot intelligence.

Primary Contact: benjamin.alt@uni-bremen.de

Preliminary Schedule

Day 1

09:45–10:00	Opening remarks
10:00–10:30	Invited Talk 1
10:30–11:30	Short Paper Session 1 (4 papers × 14 min)
11:30–11:45	Poster Lightning Talks 1 (6–7 posters × 2 min)
11:45–12:45	Lunch
12:45–13:15	Invited Talk 2
13:15–14:15	Short Paper Session 2 (4 papers × 14 min)
14:15–14:40	Poster Lightning Talks 2 (6–7 posters × 2 min)
14:40–16:00	Coffee break and Poster Session
16:00–16:30	Invited Talk 3
16:30–17:10	Plenary discussion / Q&A
17:10–18:00	Panel introduction / informal networking

Day 2

09:00–09:30	Invited Talk 4
09:30–10:30	Short Paper Session 3 (4 papers × 14 min)
10:30–11:00	Poster Lightning Talks 3 (8–10 posters × 2 min)
11:00–11:30	Invited Talk 5
11:30–12:30	Lunch
12:30–13:30	Short Paper Session 4 (4 papers × 14 min)
13:30–15:30	Poster Session and mentoring activities
15:30–16:30	Panel discussion / Q&A
16:30–17:00	Wrap-up / informal networking

Contact

For inquiries, please contact: benjamin.alt@uni-bremen.de

1st IJCAI Workshop on Safe Physical AI

IJCAI/ECAI 2026