Workshop on Embodied Intelligence for Autonomous Systems on the Horizon

Event: CVPR 2025 Nashville · Duration: 0 min · ▶ Watch on YouTube

Abstract

The CVPR 2025 Workshop on Embodied Intelligence for Autonomous Systems on the Horizon explores the critical role of embodied intelligence in advancing autonomous driving and robotics. It delves into the data pyramid, encompassing real-world, public, synthetic, and simulated data, as foundational elements for developing intelligent systems. The workshop aims to identify key challenges and opportunities in building foundation models, understanding value/reward learning, and establishing lifelong learning frameworks for automated systems. It also features an Autonomous Grand Challenge to benchmark progress in multi-modal understanding, end-to-end multi-modal learning, embodied agents, and world models, with a long-term vision towards Artificial General Intelligence (AGI).

Speakers

  • Jitendra Malik — UC Berkeley
  • Abhinav Gupta — CMU
  • Felix Heide — Princeton
  • David Crandall — Indiana University
  • Vincent Vanhoucke — Waymo
  • Antonio Loquercio — UPenn
  • Fatma Güney — Koç University

Talks (11)

  • 09:00:00 — Chonghao Sima: Opening Remarks
    • Opening remarks for the workshop on embodied intelligence for autonomous systems.
  • 09:10:00 — Fatma Güney: Scaling Up Self-Driving: Using Large Models Efficiently and Learning Rewards from Videos
    • Discussion on scaling self-driving capabilities through efficient large models and video-based reward learning.
  • 09:50:00 — Track Organizer and Winners: NAVSIM v2 End-to-End Driving Challenge
    • Presentation of the NAVSIM v2 End-to-End Driving Challenge by its organizers and winners.
  • 10:20:00 — Felix Heide: Scalable Autonomous Driving via Fully Data-driven Simulation
    • Insights into achieving scalable autonomous driving through advanced data-driven simulation techniques.
  • 11:00:00 — Antonio Loquercio: What does Embodied Intelligence mean? Lessons Learned from Drone Racing
    • Exploring the meaning of embodied intelligence with lessons drawn from the field of drone racing.
  • 13:00:00 — Jitendra Malik: Theme TBD
    • A presentation by Professor Jitendra Malik on a to-be-determined theme.
  • 13:40:00 — Vincent Vanhoucke: Foundation Models for Autonomous Driving
    • An overview of foundation models and their application in autonomous driving systems.
  • 14:20:00 — Track Organizer and Winners: World Model Challenge by 1X
    • Presentation of the World Model Challenge by 1X by its organizers and winners.
  • 14:40:00 — David Crandall: Theme TBD
    • A presentation by Professor David Crandall on a to-be-determined theme.
  • 15:30:00 — Abhinav Gupta: Theme TBD
    • A presentation by Professor Abhinav Gupta on a to-be-determined theme.
  • 16:10:00 — Speakers and Organizers: Debate on Path to Achieving Autonomy
    • A panel debate among speakers and organizers discussing the future path towards achieving full autonomy.

Key Takeaways

  • The workshop focuses on the progression from multi-modal understanding to end-to-end multi-modal systems and embodied agents, aiming for advanced world models and AGI.
  • It highlights the importance of integrating real-world data, public datasets (like YouTube, Open-X), and synthetic/simulation data to build robust embodied intelligence.
  • The Autonomous Grand Challenge 2025 includes specific tracks such as NAVSIM v2 for end-to-end driving, a World Model Challenge by 1X for humanoid robot interactions, and the AgiBot-World Challenge for manipulation capabilities.
  • Key research questions revolve around the development of foundation models, value/reward learning, and lifelong learning frameworks for autonomous systems.
  • The event features a diverse lineup of speakers from leading academic institutions and industry, fostering discussion on the future of embodied intelligence.

Methods / Models / Datasets Mentioned

  • MotionLM
  • EMMA
  • NAVSIM v2
  • AgiBot World
  • CARLA

Topics

Embodied Intelligence · Autonomous Systems · Foundation Models · Autonomous Driving · Robotics · Multi-modal Learning · World Models · Simulation · Artificial General Intelligence (AGI) · Reinforcement Learning

Notes

Open for commentary — connections to other work, critiques, follow-up reading.