The proposed LLM-BT method consists of four key modules:
Recognition: Constructs a semantic map by using a 3D object recognition algorithm to obtain information about objects in the real-time scene.
Reasoning: Employs the reasoning capability of ChatGPT to understand the information from the semantic map and user input, and generate descriptive steps of the task.
Parser: Utilizes a BERT-based LLM to extract keywords from the descriptive steps and construct an initial BT that represents the goal of the task.
BTs Update: Proposes an algorithm to dynamically expand the initial BT by adding new actions and assigning appropriate executing priorities based on environmental changes, enabling the robot to handle external disturbances.
Compared to other LLM-based methods for complex robotic tasks, LLM-BT has the advantage of adaptability, as it outputs variable BTs that can add and execute new actions according to environmental changes, making it robust to external disturbances.
The experiments on cargo sorting and household service tasks demonstrate the feasibility of the LLM-BT method, where the robot was able to adapt to various external disturbances, such as dropped objects or obstacles, by dynamically updating the BT.
To Another Language
from source content
arxiv.org
Ключові висновки, отримані з
by Haotian Zhou... о arxiv.org 04-09-2024
https://arxiv.org/pdf/2404.05134.pdfГлибші Запити