This project aims to investigate novel machine learning methods for teaching robots to move their vision/camera system more intelligently for finding objects/targets in highly occluded and unstructured environments. Deep reinforcement learning is a machine learning technique which has the potential to enable robots to learn complex tasks through high dimensional input sensors such as cameras.
This project will investigate using reinforcement learning techniques for learning robotic policies to intelligently move a camera on the end of a robot to see around occlusions in unstructured environments. This has applications in standard robotic pick and place environments but in more challenging unstructured and natural environments such as agriculture.
- Train robot policies using deep reinforcement learning for handling occlusions in unstructured environments
- Define the problem framework and reward function to solve the reinforcement learning problem.
- Program robotic arms for learning and executing policies
- A robotic system which can move its vision system around occlusions in an unstructured environment
Skills and experience
- Machine learning and statistics, preferably some understanding of reinforcement learning methods
- Python programming
- Understanding of computer vision and robot arm kinematics/dynamics
Contact the supervisor for more information.