1st Workshop on
Generalization in Autonomous Driving: Paradigms, Practice, and Public Road Demonstrations

Autonomous driving has made remarkable progress in recent years. Nevertheless, one unsolved question remains: How can autonomous vehicle (AV) systems generalize to new environments or unseen conditions?
This question can be answered from different angles: (1) Use machine learning-based software architectures, (2) use more data, (3) apply heavy simulation & real-world testing or (4) making code open-source available for joint community development. This workshop will go beyond abstract discussions by confronting participants with the real challenges of generalization in autonomous driving. We will stage a direct dialogue between two competing paradigms: (1) modular, open-source ecosystems that thrive on community-driven collaboration, and (2) data-hungry end-to-end approaches that aim to scale through massive models and datasets. What makes this workshop unique is its hands-on character: participants will experience a live, livestreamed demonstration of the TUM EDGAR autonomous vehicle driving on public roads in Munich, providing a concrete basis for in-depth discussions on robustness, safety, and scalability. Through keynotes, interactive sessions, and real-world experiments on topics like domain adaptation, sim-to-real transfer, self-supervised and continual learning, evaluation benchmarks, and software engineering practices, we will collectively ask: which paradigm — modular, end-to-end, or hybrid — can truly deliver generalization in AVs?

The workshop will feature prominent speakers, and contributions from the intelligent vehicles and mobile robotics community.
The workshop is happening in-person at ICRA 2026 in Vienna. Additionally we welcome participants to listen and contribute virtually via Zoom.

The submission phase is closed. Decisions have been communicated to authors via OpenReview. All accepted papers are invited to participate in the workshop poster session.

• 23 Submissions
• 13 Accepted papers
• 56.5% Acceptance rate
• 13 Poster invitations

*Listed papers are accepted workshop submissions. PDFs may not yet be the final camera-ready versions and will be updated.

We thank all reviewers for their time, expertise, and constructive feedback to the authors.