[Taipei, June 03, 2026] Foxlink Group Executive Assistant and SYNCROBOTIC President Freddy Kuo was invited to speak at NVIDIA GTC Taipei 2026, delivering a session titled “From Simulation to Factory Floor: Advancing Robot Mobility, Navigation, and Intelligent Decision-Making in Real-World Operations.” The presentation placed SYNCROBOTIC’s Physical AI technology stack at the center — offering a full-stack view of how Foxlink Group has built an end-to-end AI deployment pipeline, from data acquisition to multi-agent intelligent orchestration, across high-volume smart manufacturing environments.
Solving the Data Problem: Smart Gloves Capture Real-World Manipulation Data
The foundational barrier to Physical AI adoption in manufacturing is data scarcity — specifically, the near-total absence of real-world manipulation data capturing how skilled workers move, pick, and place. Mr. Kuo highlighted the pervasive issue of data fragmentation across factory environments, where the most critical training data simply does not exist as a ready resource.
To solve this, SYNCROBOTIC developed its proprietary Smart Gloves data acquisition system, which captures skilled operator hand movements with ultra-low latency and converts them directly into scalable AI training datasets — addressing the bottleneck at its source.
Redesigning Industrial AI: From Workstations to Action Primitives
SYNCROBOTIC introduces a fundamental shift in how industrial AI models are built. Rather than centering automation around fixed workstations — welding, dispensing — the company’s Physical AI architecture is built on reusable action primitives such as pick and place. This paradigm allows a single AI model to be redeployed across tasks and production lines by simply swapping end-effectors, significantly improving model generalizability and deployment efficiency.
Sim2Real Pipeline: From Digital Twin Validation to Factory Deployment
To handle the variability of live manufacturing environments, SYNCROBOTIC has established a full Sim2Real closed-loop learning system. The pipeline runs: Smart Gloves data capture → digital twin simulation and physics validation in NVIDIA Isaac Sim and Isaac Lab → algorithm testing via MuJoCo → factory floor deployment → real-world feedback for continuous model refinement. This closed loop ensures AI is production-ready before any physical deployment.
Smart Orchestration: Real-Time AI Decision-Making Across Systems
At the orchestration layer, SYNCROBOTIC integrates NVIDIA NemoClaw‘s large language model reasoning with its proprietary factory operating system, LocalMind, enabling real-time Smart Orchestration across multiple agents and systems. When production cameras detect an obstacle, AI autonomously re-routes, reassigns AGVs, and completes the task — without human intervention. This is AI functioning as a decision-maker in a dynamic industrial environment, not just an executor.
Verified Results on Foxlink’s Factory Floor
SYNCROBOTIC’s Physical AI platform has been deployed and validated across Foxlink Group’s high-volume manufacturing facilities, delivering the following quantified outcomes:
- 33% faster production ramp-up
- 25% reduction in equipment maintenance workload
- 50% decrease in quality escapes
- 2–5× increase in inspection frequency via autonomous patrol robots, with 30–50% reduction in manual inspection effort
These results are not the product of isolated optimizations — they reflect a complete technology loop, from data acquisition and simulation validation to real-time on-site orchestration. This is the core challenge SYNCROBOTIC has been building toward: AI that works on the factory floor, not just in the lab.
