Date: November 22, 2024
Location: Nancy, France (Part of the Humanoids 2024 Event)
Overview:
General-purpose robots are machines that could perform virtually any task given simple instructions (e.g., verbal instructions). Up to a few years ago, these robots were mostly considered to be out of reach, often because of the lack of "common sense" displayed by the planning and natural language processing algorithms.
Recent advances in Generative AI, including Large Language Models (LLMs), Visual Language Models, and Diffusion methods, are making this vision increasingly attainable. Models such as ChatGPT demonstrate the ability to interpret and respond to diverse queries from text or images, proposing relevant answers and plans. However, linking such models to actions or control is an open challenge.
For these advanced AI models to be effectively utilized in robotics, the robots themselves must possess general-purpose capabilities. Humanoid and quasi-humanoid robots (e.g., bimanual torsos on wheels, centaur-like configurations) are particularly well-suited for this role: their versatility allows them to perform a wide range of tasks in environments are designed for humans.
The potential synergy between generative AI and humanoid robots presents an exciting avenue for research and industry that is gaining significant attention within the scientific community, with various research teams exploring multiple approaches, both with and without training specific components, and leveraging language or visual language models.
This workshop aims to foster a comprehensive dialogue among multi-disciplinary research areas (e.g. AI, data, robotics), providing a platform to share insights, methodologies, and visions for the future of humanoid robotics powered by generative AI. This is why the program balances speakers from academia, policy, and industry. In addition, we will circulate a call for contributions for short presentations by PhD students and post-docs if their work combines Generative AI and Humanoids.
The discussions in the workshop will contribute to the development of a European industrial strategy toward general-purpose robotics within the AI, Data and Robotics Association (ADRA) that is co-organizer of the workshop.
Preliminary Program:
Each session includes time for Q&A and audience participation.
14:00 - 14:05 — Introduction
14:05 - 14:30 — Academic Presentation #1: Working with Large Language Models: Challenges and Solutions — Christophe Cerisara (LORIA, CNRS)
14:30 - 14:45 — Policy Presentation: Technology Roadmap for Europe and Beyond — Catherine Simon
14:45 - 15:00 — Industry Presentation: Replacing Manual Work in Manufacturing with Safe, Dexterous, and Intelligent Robots — Thibauld Jongen (Founder, Generative Robotics, director AI, Data, and Robotics Association)
15:00 - 15:30 — COFFEE BREAK
15:30 - 16:15 — Academic Presentation #2: Connecting Generative AI and Robotics — Edward Johns (Imperial College)
16:15 - 16:30 - Statements by the panel guests - Francesco Nori - Jérome Monceaux - Camille Croze
16:30 - 17:45 — Panel Discussion — Interactive discussion between invited speakers and the audience - The Roadmap of Generative AI for Robotics
17:45 - 17:50 - Wrap-up
- Moderator: Francesco Ferro (PAL Robotics)
- Guests:
- Edward Johns (Imperial College)
- Christophe Cerisera (CNRS)
- Thibauld Jongen (Generative Robotics, Director AI, Data, and Robotics Association)
- Catherine Simon (SGPI)
- Francesco Nori (Google deepmind)
- Jérome Monceaux (Enchanted Tools)
- Camille Croze (Tplusone)