The construction of unmanned laboratories can effectively promote the efficiency of multi-type drug synthesis and new drug development. This paper aims to propose an autonomous decision-making framework for composite robots to address semantic workflow consistency in long-horizon drug synthesis tasks, thereby assisting researchers in accurately performing experimental operations.
This paper proposes a hierarchical autonomous decision-making framework for experimental robots based on multi-level tasks and skill knowledge graph constraints. First, by constructing a staged task model and historical behavior memory features, it enhances the policy model’s perception of task states and action dependencies, improving precise decision-making of task-level action sequences. Meanwhile, it proposes a semantically driven scoring mechanism, combined with heuristic search, enabling efficient retrieval of subtask skill chains under complex constraints and improving the robot’s autonomous operation capability in drug synthesis tasks.
Experimental results show that the proposed method can perform fast decision-making for behavior sequences of complex long-horizon drug synthesis tasks under knowledge graph constraints, achieving an average step prediction accuracy of 93.2%. In addition, by integrating a subtask skill search method, it enables fast retrieval of skill chains required for subtask nodes, improving the capability of composite robots to complete complex sequential tasks, with a maximum single-task success rate improvement of 32.4% compared to baseline models.
This study presents a knowledge-graph-constrained hierarchical autonomous decision-making framework for robots, which effectively enhances dynamic decision-making capabilities in complex, long-horizon drug synthesis scenarios with strong behavioral dependencies, thereby ensuring the effectiveness of the drug synthesis process.
