Title:Robot Imitation from Video Demonstration

Abstract:This paper presents an attempt to replicate the robot imitation work conducted by Sermanet et al., with a specific focus on the experiments involving robot joint position prediction. While the original study utilized human poses to predict robot joint positions, this project aimed to achieve robot-to-robot imitation due to the challenges of obtaining human-to-robot translation data. The primary objective was to provide a neural network with robot images and have it predict end-effector positions through regression. The paper discusses the implementation process, including data collection using the open-source RoboSuite, where a Python module was developed to capture randomized action data for four different robots. Challenges in data collection, such as oscillations and limited action variety, were addressed through domain randomization. Results show high testing error and unsatisfactory imitation due to overfitting, necessitating improvements in the project.