Modeling Uncertainty in Robotic World Models

Recent advances in vision-language foundation models and video generation have made learned world models an increasingly important tool for robotic task planning, policy simulation, and reward modeling. Yet real-world robots must operate under uncertainty that many current world models do not adequately represent: noisy sensors, partial observability, imprecise actuation, distribution shift, and unpredictable humans or other agents.

Although related communities have studied uncertainty through belief-space planning, probabilistic robotics, conformal prediction, goal inference, safe autonomy, and uncertainty quantification, it remains unclear how these ideas should scale to the high-dimensional latent spaces used by modern learned world models. This workshop will bring together researchers from robot learning, planning, probabilistic robotics, safe autonomy, uncertainty quantification, and human-robot interaction to address that gap.

We aim to clarify which uncertainty representations fit which robotic settings, how uncertainty propagates into downstream planning and reinforcement learning, and which benchmarks and metrics are needed to measure progress. Through invited talks, poster presentations, and roundtable discussions, our goal is to move robotic world models beyond expressiveness alone toward reliable uncertainty awareness for deployment in unstructured real-world environments.

{{ d.label }}

{{ d.value }}

{{ g.label }}

{{ p.t }}

Areas of Interest

Topics include, but are not limited to:

{{ t.n }} {{ t.text }}

{{ s.n }}

{{ s.name }}

{{ s.affiliation }}

Tentative Schedule

Room and location: TBD. All times local to Austin (CST).

{{ s.time }}

{{ s.event }}

{{ p.name }}

{{ p.affiliation }}

{{ p.name }}

{{ p.affiliation }}

Call for Reviewers