

Physical AI in Production: From Models to Real-World Robots
Robotics models can look convincing in demonstrations. Deploying them around people is a different challenge: real-time decisions, safety, changing environments, and the engineering needed to keep physical systems reliable.
Join Tokyo AI (TAI) for an evening of three technical talks on what it takes to move Physical AI from research and demos into working robots. We will examine practical architectures, deployment constraints and lessons from real-world systems.
When: Tuesday, December 1, 2026, 18:00–21:00 JST (provisional time).
Where: Tokyo — venue to be confirmed.
Arjun Prakash has provided a draft talk for the second slot, pending his updated abstract and final event confirmation. One additional speaker will be announced.
Agenda (provisional)
18:00–18:30 — Arrival and networking
18:30–19:00 — Talk 1: Challenges in Deploying Physical AI on Robots — Alisher Abdulkhaev, Kanaria Tech
19:00–19:30 — Talk 2: Coming to Grips with Autonomous Research — Arjun Prakash
19:30–20:00 — Talk 3: To be confirmed
20:00–21:00 — Discussion and networking
Talks
Challenges in Deploying Physical AI on Robots
Speaker: Alisher Abdulkhaev, Co-founder & CTO, Kanaria Tech
Abstract
World models and VLAs make robots look solved in demos. Deployment is a different problem: a learned model is too slow to be the only thing between a robot and a person, it will sometimes be wrong, and every robot and site is different. Alisher walks through KRM, Kanaria's three-layer navigation stack, and the deployment problem each layer exists to solve.
Bio
Alisher Abdulkhaev is co-founder and CTO of Kanaria Tech, a Tokyo-based deep-tech startup building KRM, a navigation system for robots that move among people. Before Kanaria he was Head of AI and Data Science at ROMS and worked as a computer vision engineer at PKSHA Technology and Rakuten. He carried out PhD research on video action recognition at the University of Tsukuba and served as a board director of Machine Learning Tokyo, a non-profit AI community of more than 14,000 members.
Coming to Grips with Autonomous Research
Speaker: Arjun Prakash, PhD Candidate, Department of Computer Science, Brown University; Summer Robotics Researcher at Capeki.
Abstract
Advances in AI now mean that autonomous research is quickly becoming a viable research methodology. The implicit assumption that research is a fundamentally human act no longer holds. To address this change, we are running the Autonomous Machine Learning Workshop at NeurIPS, where an autonomous agent made a decisive contribution to the paper’s primary result.
I will present preliminary findings from the Workshop on how ML researchers are approaching autonomous research, and where they remain undecided. I will also present my own results applying autonomous research to robotics, where I have had success directing agents to dramatically improve model inference speed on unconventional hardware.
Bio
Arjun is a PhD candidate at Brown University, where he researches reinforcement learning and robotics. Arjun's research agenda involves understanding how neural networks can be trained indefinitely for robotic control in changing environments. He has recently become interested in autonomous research.
Talk 3
Details coming soon.
Organizers
Tokyo AI (TAI) brings together Japan's AI engineers, researchers, founders and technology leaders through technical talks and community events. Cedric will lead this event onsite.
Supporters
Foundry Labs K.K. is a Tokyo-based AI systems integrator and solutions provider, delivering enterprise AI systems from strategy and engineering through deployment and operations.
About TAI
Tokyo AI (TAI) is an international AI community in Japan with more than 7,000 members. Our goal is to connect people building AI in Japan with the global ecosystem.
Privacy Policy
We will process your email address for event-related communications and ongoing newsletter communications. You may unsubscribe from the newsletter at any time. Further details are available in our Privacy Policy.