Published on:
2 November 2023
Primary Category:
Robotics
Paper Authors:
Andrea Tagliabue,
Kota Kondo,
Tong Zhao,
Mason Peterson,
Claudius T. Tewari,
Jonathan P. How
REAL enables adaptation at both low-level control and high-level mission planning in aerial robots
A large language model is queried online to choose actions based on natural language logs and error codes
The method is demonstrated on a real multirotor, improving tracking under uncertainties
The large language model can trigger emergency landings based on detecting issues in logs
The initial prompt requires minimal knowledge of the robot's dynamics or mission
Using large language models to enable adaptation and resilience in autonomous aerial robots
This paper presents REAL, a method to leverage large language models' prior knowledge and reasoning abilities to enable online adaptation and decision-making across different components of an autonomous aerial robot's software stack. REAL allows the aerial robot to improve its trajectory tracking and make mission-critical decisions by querying a large language model, requiring only a simple initial prompt and no explicit dynamics modeling.
Reasoning for Robot Control with Language Models
Embodied AI: Bringing Language Models Into the Physical World
Fine-tuning language models for robot programming assistance
Large language models for autonomous vehicle decision-making
Language reasoning for robot navigation
Using AI to Help Robots Plan Tasks and Handle Unexpected Situations
No comments yet, be the first to start the conversation...
Sign up to comment on this paper