VicOne Showcases Rthena Real-Time Robotics Cybersecurity at GTC Berlin
AI assistants · Enterprise AI · Smart operations
At GTC Berlin, VicOne presented physical AI robotics security solutions, featuring pre-deployment tool Radeis and runtime defense model Rthena. Built on NVIDIA Nemotron, Rthena operates at the edge using real-time reasoning and prompt patching to intercept rogue commands while maintaining continuous operation through cloud fleet defense.
VicOne demonstrated cybersecurity solutions for physical AI robots at NVIDIA GTC Berlin. CEO Leo Cheng noted that physical AI introduces uncertainties such as data poisoning, model hallucinations, and sensor misinterpretations, extending cyber threats into physical safety risks. VicOne's framework includes Radeis, which integrates NVIDIA Isaac Sim to simulate attack scripts in digital twins prior to deployment, and Rthena, an edge defense model trained on NVIDIA Nemotron for runtime monitoring. Rthena can be embedded at the factory or retrofitted by enterprises under an 'edge real-time response, cloud fleet defense' model. When robots face spoofed low-battery signals or voice injection on operating systems like ROS2, Rthena detects inconsistencies and generates prompt patches to bypass risky instructions, avoiding complete shutdowns during inspection or delivery missions. Incident logs are subsequently reported to cloud security operations to update fleet-wide policies.
Key takeaways
- VicOne introduced Radeis, combining NVIDIA Isaac Sim for digital twin red-team attack simulation before deployment.
- Trained on NVIDIA Nemotron, Rthena operates at the edge to cross-check sensor feeds and paths against operational baselines.
- Against threats like voice injection, the system generates real-time prompt patches to bypass risky commands without halting tasks.
- The setup combines edge rapid mitigation with cloud fleet defense, routing anomaly logs to refine fleet-wide security policies.
Source：科技報橘 Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.