Hierarchical Robot Task Framework Combining Knowledge and Learning

Hierarchical Robot Task and Motion Planning Framework Combining Knowledge Bases and Learning Models

This technology is a hierarchical framework that receives task and object names as input to generate task sub-goals using a learning model, converts them into robot-level task actions through object knowledge and PDDL-based graph search, and generates execution motion plans by utilizing a motion knowledge base and primitive actions.

Existing task and motion planning methods face challenges with large search spaces when performing complex, long-horizon tasks. Furthermore, limitations in symbolic task planning make it difficult to guarantee success rates in diverse, heterogeneous robot environments and restrict the generation of flexible plans.

This technology introduces a feedback loop that decomposes task goals using learning-based models and corrects infeasible plans through simulation, while ensuring feasibility via a knowledge database. It can be applied to autonomous tasks in service robots and smart factories, significantly improving the success rate of complex, long-term operations.

‍

‍

Key Features:

• Step of generating a robot task plan through a robot task planning framework that combines knowledge and learning
• Step of deriving a robot motion plan to control the robot based on the generated task plan
• Configuration that generates sub-goals from task and object names using a learning model trained on object state changes
• Configuration that constructs a task plan for the generated sub-goals using knowledge stored in an object knowledge base