What they built

NITROGUARD

NitroGuard is a real-time AI safety gateway for autonomous mobile robots (AMRs) operating in dynamic factory environments, addressing an emerging challenge in physical AI and industrial automation. As large language models are increasingly used for natural-language mission planning in embodied AI systems, a structural gap remains: LLMs lack deterministic spatial awareness and cannot inherently guarantee collision-free execution. Existing industrial safety standards, such as ISO 10218:2025, were built for conventional control architectures and have not yet been extended to account for AI-generated motion intent. NitroGuard addresses this by acting as an MCP-native interception layer between AI-driven planning and physical actuation. Every movement command proposed by the LLM is evaluated by a Control Barrier Function (CBF) safety engine before execution, mathematically constraining the trajectory outside defined hazard boundaries in real time. The language model never holds direct write access to the robot. Mission planning is grounded in live MCP Resources — factory layout, hazard map, robot state — before a target is proposed. The raw versus safety-corrected trajectory is rendered live in 2D/3D, then dispatched to a MuJoCo physics simulation for execution. Built on the full NitroStack framework: Tools for mission execution and emergency stop, Resources for environmental grounding, a Prompt template for safe-navigation workflows, Guards and rate-limiting on execution, and an interactive trajectory-viewer Widget. Intended for robotics teams integrating LLMs into control loops, industrial manufacturers adding natural-language interfaces, and integrators needing auditable safety assurances as AI-robot regulation develops.