Design and development of a quadruped robot for Human assistance

Abstract

This thesis proposal focuses on enhancing the intelligence and versatility of quadruped robots through the integration of Large Language Models (LLMs) and advanced control algorithms, including Model Predictive Control (MPC) and Whole Body Control (WBC). These integrations aim to enable the robot to perform complex locomotion tasks while responding intelligently to human commands in various environmental settings. The mechanical design and control framework of the proposed quadruped robot are explored, emphasizing the implementation of multi-rigid-body models to optimize control tasks hierarchically. The inclusion of LLMs is intended to allow the robot to interpret and execute human instructions more effectively, thereby enhancing its capability to interact naturally and adaptively with humans. Anticipated applications of this research include roles in disaster response, material transportation, and space exploration, where the robot’s ability to navigate challenging terrains and perform tasks autonomously could be crucial. The thesis will detail the progression from initial design phases through to practical implementation and testing, including the development of LLM architectures, data preparation, and model training methods. The ultimate goal is to produce a quadruped robot that not only meets but exceeds current robotic capabilities in interacting and assisting humans in complex environments.