RISE 2026 – Reasoning in Scalable Embodied Intelligence (RISE) Workshop

Introduction

The 1st edition of the Reasoning in Scalable Embodied Intelligence (RISE) workshop at ICML 2026 focuses on the critical role of reasoning in transitioning embodied AI agents from controlled laboratories to open-world deployment in robotics, autonomous driving, and interactive physical environments.

While advances in robot learning have significantly improved perception and control, robust real-world performance increasingly depends on reasoning: the ability to interpret context, infer latent structure, anticipate long-horizon consequences, and adapt to previously unseen situations.Despite rapid progress, key questions remain regarding how reasoning should be represented, learned, and integrated with world models to move toward general-purpose embodied agents.

RISE aims to bridge traditionally separate communities (machine learning, robotics, computer vision, and autonomous systems) around the shared challenges of learning to reason in real-world environments. We emphasize learning-based reasoning grounded in perception and action, exploring multimodal foundation models, generative simulation, and reinforcement learning.

The workshop invites researchers and practitioners to:

Investigate visual, linguistic, symbolic, or latent reasoning mechanisms critical for embodied tasks.
Develop architectures and world models for dynamics, causality, and counterfactual reasoning.
Address the challenges of out-of-distribution generalization and long-tail scenarios through reasoning-aware learning.

Topics of Interest

Foundations of embodied reasoning: visual, language-based, symbolic, and latent mechanisms
Structured, programmatic, and chain-of-thought (CoT) reasoning grounded in physical interaction
Learning to reason from interaction via reinforcement learning, imitation learning, and planning
Reasoning for out-of-distribution (OOD) generalization and long-tail scenarios
Hierarchical, modular, and compositional architectures for embodied reasoning and control
World models for dynamics, causality, counterfactual reasoning, and policy learning
Generative models for imagination, simulation, and long-horizon planning
Vision-Language-Action (VLA) models and multimodal foundation models for embodied intelligence
Language grounding, instruction following, and task planning in real environments
Robotics and autonomous driving systems enabled by embodied reasoning
Datasets, benchmarks, evaluation protocols, and real-world deployment studies

Schedule (Tentative)

Time	Session
08:00 – 08:10	Opening Remarks – Welcome & Workshop Overview
08:10 – 08:50	Keynotes 1 - Prof. Hang Zhao (Tsinghua University) Keynote
08:50 – 09:30	Keynotes 2 - Prof. Sherry Yang (NYU & DeepMind) Keynote
09:30 – 09:50	Paper Oral Presentation I Oral
09:50 – 10:10	Coffee break
10:10 – 10:50	Keynotes 3 - Prof. Yilun Du (Harvard University) Keynote
10:50 – 11:10	Paper Oral Presentation II Oral
11:00 – 12:00	Poster Session Posters
12:00 – 13:00	Lunch Break
13:00 – 14:00	nuReasoning Challenge Session: introduction, winner awards, and lessons learned Challenge
14:00 – 14:40	Keynotes 4 - Prof. Chelsea Finn (Stanford University) Keynote
14:40 – 15:00	Best Paper Award & Oral Presentation III Oral
15:00 – 15:20	Coffee Break
15:20 – 16:00	Keynotes 5 - Prof. Serget Levine (UC Berkeley & Physical Intelligence) Keynote
16:00 – 16:40	Panel DiscussionPanel
16:40 – 17:00	Closing Remarks & Group Photo

Final schedule, room allocation, and speaker order will be announced closer to the workshop date.

Confirmed Keynote Speakers

Dr. Balajee Kannan

Motional

Prof. Hang Zhao

Tsinghua University

Prof. Sherry Yang

NYU & DeepMind

Prof. Yilun Du

Harvard University

Prof. Chelsea Finn

Stanford University

Prof. Sergey Levine

UC Berkeley & Physical Intelligence

Paper Track

RISE 2026 invites high-quality contributions on reasoning in embodied intelligence, and related topics outlined above.

Submit Now

Paper Type: 4-page extended abstracts or 8-10 page versions of recently published work
Included in Proceedings: No (non-archival)
Submission Deadline: May 1, 2026 (23:59 PST)
Notification to Authors: May 10, 2026
Camera-Ready Deadline: May 20, 2026

Paper Awards

🏆 Best Paper Award: $1,000
🏆 Best Paper Runner-Up Award: $500

Challenge Awards

🏆 nuReasoning Challenge Winner Award: $10,000
🏆 nuReasoning Challenge Runner-up Award: $5,000
🏆 nuReasoning Challenge Innovation Award: $2,000

nuReasoning Challenge

The nuReasoning Challenge aims to develop autonomous driving agents that can robustly handle rare interactions and complex causal dependencies, going beyond average-case perception and planning. This open dataset and benchmark is inspired by Motional's nuScenes and nuPlan datasets and the NAVSIM end-to-end driving benchmark, but is specifically targeted at reasoning-intensive, long-tail driving scenarios

Challenge Objectives

Reasoning-Centric Planning: Given sensor observations and context, agents must plan safe, human-compatible trajectories while outputting explicit reasoning signals, such as meta-actions.
Long-Tail Robustness: Evaluation focuses on performance during rare interactions, including foreign object debris, occlusions, and uncommon causal dependencies.
Human-Verified Logic: The challenge utilizes the nuReasoning Dataset, which features realistic driving logs enriched with high-quality, human-verified reasoning annotations.

Top-performing submissions will be invited to present their findings at the workshop. Cash prizes will be awarded to the winners of the nuReasoning challenge. All materials, including the dataset, code, and evaluation scripts, will remain publicly available after the workshop to support continued research.