Vision for Action: Long-term Visual Prediction for Vision-based Robot Manipulation

Supervisor: Dr Changjae Oh

Human vision systems can support action to interact with objects in physical environments. Similarly, computer vision techniques facilitate a robot, or an AI agent, to deal with the real-world environment. While recent researches in machine learning (ML) based robot manipulation focus on investigating ML models to generate actions from raw visual input, they have shown less attention to explicitly integrating computer vision techniques. The goal of this project is to investigate computer vision techniques together with ML-based robot manipulation pipeline. Specifically, this project will investigate a predictive cognition model to enhance the capability of learning robot manipulation tasks. Strong programming skills and background in computer vision are required.