Building Information Models to Robot-Ready Site Digital Twins (BIM2RDT): An Agentic AI Safety-First Framework

Safety is embedded throughout the framework design: BIM2RDT explicitly prioritizes safety during autonomous robot operations on construction sites by integrating real-time sensing data with pre-planned models, rather than treating safety as an afterthought. This "safety-first" approach means robots can continuously verify actual site conditions against expected conditions before executing tasks.

AI systems need semantic understanding, not just pattern matching: The framework's key innovation—Semantic-Gravity ICP—shows that large language models can improve AI accuracy by reasoning about object semantics (what things are and how they should be oriented) rather than relying solely on geometric pattern matching. This reduces errors from the algorithm getting "stuck" in incorrect solutions.

Real-time feedback loops enable continuous improvement: The system creates a cycle where robots collect visual data that updates the digital twin, which then optimizes future robot paths and operations. This means autonomous systems can learn from actual site conditions and adapt, rather than rigidly following pre-programmed plans that may not match reality.

Multiple data sources create more reliable AI decisions: By combining three independent data streams—design models (BIM), sensor networks (IoT), and robot observations (visual-spatial)—the framework reduces the risk that AI systems will make decisions based on incomplete or outdated information, which is critical for safety-sensitive construction environments.