WIP: Embodied Intelligence : Leveraging LLMs for Decision Making and MoCap for Gait

Humanoid robotics is a rapidly growing field in artificial intelligence, focusing on developing robots that can interact with human environments. Our ongoing research aims to enhance humanoid robotics by integrating Large Language Models (LLMs) to improve robots' decision-making abilities. Our proposed architecture integrates LLMs with humanoid robots to facilitate decision making and seamless operations. We are exploring the potential of LLMs to enhance complex decision-making processes within humanoid frameworks. To ensure effective physical interaction, we have employed traditional imitation learning techniques to generate gait patterns based on a comprehensive motion capture dataset. By integrating these diverse components, we aim to create a sophisticated humanoid robot capable of intelligent and adaptable behavior. Our research has utilized Reinforcement learning’s Stable Baseline3’s algorithms, including Advantage Actor Critic (A2C), Deep Deterministic Policy Gradient (DDPG), Soft Actor Critic (SAC), and Twin Delayed DDPG (TD3) to date.