From Language to Deployment: Offline Optimization and Ontology-Guided Behavior Tree Generation for Transparent Robot Applications
Ruichao Wu, Jiwei Pan, Mohamed Youssef, Bjoern Kahl, Werner Kraus, Andrey Morozov
AI summary
Problem
Current language-driven robot programming lacks the interpretability, structural validation, and robust parameter tuning required for reliable deployment in safety-critical domains.
Approach
The framework parses natural language into modular Behavior Trees, validates their control and data flows against an ontology, and systematically optimizes task parameters offline using Monte Carlo Tree Search guided by LLM priors.
Key results
- Modular Behavior Tree composition from natural language
- Ontology-guided dual-graph validation with automated repair
- Offline Monte Carlo Tree Search optimization of interdependent parameters
- Successful sim-to-real deployment and cross-platform portability
Why it matters
It enables non-experts to reliably program industrial robots from natural language while guaranteeing the transparency and correctness required for real-world deployment.
Abstract
Automatically generating robot applica- tions from natural language promises to lower the barrier to automation, but remains difficult in domains that demand reliability and transparency, such as industrial assembly or collaborative manipulation. End- to-end policies and large language model (LLM)-based planners can map instructions to robot behaviors, but they often lack interpretability and provide limited assurance of correctness. We present a framework that composes applications from modular, application- independent atomic skills expressed as Behavior Trees (BTs). BTs are constructed and validated against an ontology-level dual graph to enforce control-flow and data-flow consistency before execution, ensuring transparency and structural correctness. Application- level parameters are optimized offline in simulation using Monte Carlo Tree Search guided by LLM-derived priors. Rather than serving as a runtime optimizer, this process systematically explores interdependent parameters, producing a dataset of reliable parameteri- zations. This dataset could later be used to train gating mechanisms for online adaptation. The framework is validated in a physical robotic setup, demonstrating transparent and consistent offline generation of de- ployable applications, and laying the foundation for adaptive, real-time systems.