The One About Robots Beyond the Lab
Wed, 28 Oct 2026, 15:00
Singapore, Singapore
Free
Originally organized by Lorong AI. What does it take for AI-powered robots to work in the real world? From robots that can reason across different tasks, to models that translate what they see and understand into actions, AI is expanding what robots can do beyond controlled environments. But building robots that can operate reliably in the real world requires more than better models. They need to reason and plan, remember what has happened, respond to changes around them, and act reliably despite constraints such as latency, connectivity, and uncertainty. Together with IMDA, we are bringing you an AI Wednesdays × Technical Sharing Session edition focused on AI and robotics, exploring the models, systems, and infrastructure shaping how intelligent robots are built and deployed beyond the lab. More About the Sharings Soujanya Poria (Associate Professor, NTU) will share on “Beyond Better Actions: What Generalist Physical Agents Still Need.” What will it take to build robots that can perform many different tasks across different environments? Soujanya will argue that the challenge isn't simply collecting more robot data. Large language and multimodal models can already reason and plan, but physical agents must coordinate that slower reasoning with fast control, remember what happened earlier, and know when they need more evidence before acting. He'll explore this through three pieces of work: GE-Act 2.0, a world action model pretrained at scale; MemBodied, which adds persistent memory to vision-language-action models; and RoboQuest, a long-horizon benchmark that separates failures in evidence gathering, interpretation, planning, execution, and verification. Together, these point towards a different challenge for generalist physical agents: coordinating fast control, persistent memory, and higher-level planners that know when to gather more evidence, replan, and check their own work. Raphael Yee (Director, Griffin Labs) will share on “Griffin Alpha.” Raphael will share how Griffin Labs built Griffin Alpha, an open-weight vision-language-action (VLA) model developed in Singapore, and why the team chose to open-source it. He'll unpack how the model was developed and share results from Alpha-S, which matches π0.5 on LIBERO and ranks first among VLA models on RoboTwin 2.0 under changes in lighting, clutter, and backgrounds. … Venue: Lorong AI @ One-North, Singapore, SG. Registration is handled by the original organizer. Original event and organizer: https://luma.com/q3lcmn8i