Robotics & World Models Reading Club 07: Learning to Dream: World Models, Imagination, Path to Foundation Models for Control — San Francisco

Join us for Robotics & World Models Reading Club 07 in the heart of San Francisco! This high-energy reading group is curated for AI researchers and innovators working at the cutting edge of robotic world models, WAMs, and embodied intelligence.

🧠 About the Reading Club

In our previous gatherings, we've welcomed brilliant minds from esteemed organizations such as Boston Dynamics, Google, NVIDIA, Stanford, UC Berkeley, CMU, and Tesla, among others.

Hosted by: Junfan Zhu & Aurora Feng
Sponsored by: Neural Motion
Explore the universe of cross-embodiment data representation layers for embodied AI with us!

📚 Core Theme of Club 07

From PlaNet to DreamerV4: We will delve into constructing foundation models for control through the power of imagination.

Recent advancements suggest a revolutionary pivot from direct environmental interaction to learning through imagination. The Dreamer series exemplifies this transition, training policies within a learned simulator to provide robust control from latent world models.

This session traces the evolution from PlaNet through DreamerV1–V4, showcasing a shift towards large-scale generative models that blend diffusion transformers and reinforcement learning. We aim to foster an intuitive understanding of learning through dreaming and its implications for robotics. We will also explore challenges like long-horizon consistency and grounding.

📝 Pre-Readings

1. Foundations of World Models - The foundational idea of learning compressed latent simulators for control.
2. PlaNet: Latent Planning Begins - Early demonstrations of latent space planning.
3. Dream to Control - A shift from planning to actor-critic learning entirely in imagination.
4. Mastering Atari with Discrete World Models - Stabilizing the learning process at scale.
5. Mastering Diverse Domains through World Models - Establishing general-purpose learners across varying domains.
6. Training Agents Inside of Scalable World Models - Innovations in generative foundation models for control.

🍽 Agenda

• 2:00 PM - Doors Open & Social (Enjoy food, beverages, and unlimited strawberries 🍓!)
• 2:30 PM - Keynote by Ahmet Şemi ASARKAYA from Agility Robotics (Online access via Zoom to be announced)
• 3:00 PM - Open-floor discussion (Share your spotlight papers and insights)

📍 Location

Downtown San Francisco
Registration required to view the address.
Register Now

🔗 Join the Community

Connect with us on Discord: Join Here
Follow Saturday Robotics for updates: Twitter

🚨 Logistics

Spaces are limited, please check in by 2:00 PM. The keynote will begin promptly at 2:30 PM. Late check-ins may not be accommodated due to the high volume of inquiries.

Looking forward to an inspiring discussion! 🌟