IEI-TIA: Industrial Embodied Intelligence Trustworthy Interpretable Agent for Robotic Long-Horizon and Repetitive Tasks

With the rapid advancement of generative artificial intelligence, embodied intelligence is increasingly being integrated into robotic tasks within smart manufacturing. Industrial operations typically feature long-horizon sequences and demand high precision. However, although embodied models, particularly Vision-Language-Action (VLA) models, have shown significant promise in autonomous control with superior generalization capabilities, their inherent closed-box nature lacks the transparency and trustworthiness essential for manufacturing tasks. To address this challenge, we propose the Industrial Embodied Intelligence Trustworthy Interpretable Agent (IEI-TIA), a high-level trustworthy supervisory agent designed for industrial sorting and palletizing tasks. Rather than directly generating low-level control commands, IEI-TIA monitors robotic execution, diagnoses possible failure types at key operational stages, and provides corrective next-step instruction via invoked skills, thereby improving the reliability and interpretability of long-horizon and repetitive robotic tasks. An alignment-enhanced parameter-efficient finetuning method is proposed utilizing paired vision-language data to enable robust trustworthiness identification across various possible robotic failure types in different subtasks in sorting and palletizing. Additionally, we construct a specialized dataset and establish benchmarks for industrial embodied intelligence behavior diagnosis and instructions. Experimental results demonstrate that the fine-tuned trustworthy agent achieves 93.7% diagnosis accuracy for various task failures, representing a substantial improvement of approximately 38% over baseline general-purpose large models. By integrating the agent's step-by-step operational instructions, the success rate of robotic long-horizon and repetitive tasks is improved by 14.0%. Note to Practitioners - This paper addresses a critical problem in industrial automation, the lack of interpretability and transparency of AI-driven robotic control in multi-stage and long-horizon tasks. In high-precision environments such as sorting or palletizing, AI closed-box models, including Vision-Language-Action (VLA) models, often fail to provide safety guarantees required for manufacturing tasks, as minor execution errors can rapidly propagate and affect the entire sequence negatively. Therefore, we propose IEI-TIA, an autonomous supervisory agent that contributes to the shift from open-loop task execution to closed-loop trustworthy control with capability of providing diagnosis and instructions. Beyond identifying potential robotic failures like off-center grasps or misplaced parts, the agent also provides actionable next-step instructions that allow the embodied intelligence system to self-correct before errors escalate. A key practical advantage is that our industrial alignment-enhanced parameter-efficient finetuning (fine-tuning) method empowers the agent utilizing vision-language models with high failure-identification accuracy 93.7%, and corrective closed-loop logic without the need for retraining. The corresponding dataset and benchmarks offer a standardized framework for practitioners to evaluate and deploy interpretable embodied intelligence within existing automation pipelines. By implementing this agent, when using embodied intelligence models in manufacturing, factories can significantly reduce downtime caused by errors or mistakes made by untrustworthy AI robots, to achieve more robust and autonomous control in long-horizon robotic tasks.