Acryl, an artificial intelligence (AI) specialized company, announced on [date] that it has successfully integrated its unified AI platform 'Jonathan' with robotic systems. Building on this achievement, Acryl plans to expand the Jonathan ecosystem from software-centric applications into the Physical AI domain, including robots and autonomous systems.
This project was
conducted in collaboration with Professor Woo Hong-wook's research team at
Sungkyunkwan University. Professor Woo's team is a leading research group in
Korea's Physical AI field and recently demonstrated its research capabilities
by being selected as one of 10 teams to receive support for AI agent
development in the '2026 AI Co-Scientist Challenge Korea' organized by the
Ministry of Science and ICT. Additionally, the research team published 12
papers on Physical AI at major AI conferences including NeurIPS throughout
2025, maintaining active research activities in academia. Based on these
research achievements, Acryl plans to advance Jonathan's Physical AI
technology.
Unlike
text-based generative AI, Physical AI is a field that requires simultaneous
perception, judgment, and action in real physical environments, with complex
multimodal data processing, ultra-low latency inference, and real-time control
being key competitive advantages. The industry emphasizes sophisticated
training of Vision-Language-Action (VLA) models and stable real-time
operational capabilities as success factors for Physical AI.
Through this
integration, Acryl is credited with comprehensively implementing core elements
necessary for Physical AI commercialization by connecting Jonathan's
'full-stack' structure—spanning from ▲data preprocessing ▲model fine-tuning
▲infrastructure optimization—to robot operating environments.
Jonathan's
'FlightBase' is a data processing foundation designed to efficiently refine and
process unstructured multimodal data such as video, sensor, and behavioral data
required for VLA model training. It transforms large-scale data into
high-quality training data to help robots perceive physical environments more
precisely, and is particularly optimized for preprocessing processes needed for
VLA fine-tuning, providing a foundation for robots to quickly adapt to specific
tasks and environments.
'AgentBase'
supports VLA model fine-tuning, helping optimize general-purpose models for
industrial sites and service environments. In particular, by applying
'Multi-Agent Pipeline' technology, it enhances the feasibility of Physical AI
by implementing a structure where roles are distributed and collaboration
occurs during complex mission execution.
Additionally,
'GPUBase' applies ultra-low latency infrastructure optimization technology to
secure the real-time performance required for robot control. Acryl utilizes
'Parenting LLM' technology to supervise the VLA model's inference process,
thereby increasing accuracy and controlling inference time to not exceed
certain deadlines, enabling robots to make decisions within specified
timeframes. Furthermore, by applying Traffic Differentiation technology, it is
designed to prioritize inference traffic even in network congestion situations,
ensuring that central server decisions are delivered to robot control actuators
without delay.
Park Wae-jin,
CEO of Acryl, stated, "Physical AI's core lies not just in computational
capability, but in a structure connected like an organism from data flow to
real-time physical control. Jonathan is a platform where data (FlightBase),
intelligence (AgentBase), and infrastructure (GPUBase) are organically
combined, and will establish itself as the standard operating system (OS) for
the Physical AI era." <END>