Introduction

Since my childhood, I have witnessed the increasing random littering and pollution in nature reserves, along with the acceleration of tourism in my home country of China. However, the currently adopted solution for this problem, which entails cleaners going around the nature reserves for the purpose of searching and collecting garage, however the purpose of searching and collecting garage is highly inefficient. Moreover, travelling in different developing and developed countries as I grew up, I found that the random litter disposal is a common issue related to education, human resource management, economy, and welfare. In order to achieve effective waste management strategies, understanding the litter distribution pattern and interpreting visitor’s littering behavior is essential. Artificial Intelligence (AI), which employs machine learning and probabilistic reasoning theories, can be used to learn the distribution of litters in nature reserves and generalize the pattern, thereby shedding light on this problem.

Driven by this dream, I chose AI and Robotics as my research areas when I started undergraduate education in UW-Madison in 2011. Since January 2014, under the supervision of Professor William A. Sethares, I have been working on my four-semester independent autonomous robot research project, “Environmentally Friendly Robot: Protecting Natural Reserves against Random Littering”.

This thesis contains two major chapters that detail my study and major accomplishments in the following aspects: Simultaneous Localization and Mapping (SLAM), Image Segmentation and Object classification. The final proposed Hybrid 2D-RGBD SLAM, which combines both 2D and RGBD measures and uses each when it works best, aims to facilitate robots path planning and autonomous navigation in more dynamic, unstructured unknown environment. The Image Segmentation with One-Shot Texture Learning and Refinement Process, computes the geometric and color features, performs texture analysis and energy minimization for optimal solution. It also allows robot to learn about the texture of new environment in order to achieve a more intelligent, efficient segmentation system. In terms of efficiency, in general cases according to testing, it only requires one iteration of learning and refinement. While the study in object classification is still in progress, the current learning patterns sheds light on the classification of garbage among the natural environment—it turns out that it may not just be a problem of distinguishing “garbage vs. non-garbage”, but also “natural vs. non-natural”.

Considering the robot’s learning potential in mapping, image segmentation and classification as well as handling uncertainties in unknown environment, in the future, I wish to transform the current project into others that may contribute to education, water scarcity and many more…I also wish that the developments in robot, can be introduced as new factors to many sustainability aspects such as the long term environmental equilibrium. When thinking about the future, I always wonder: Can robots awaken people’s mind to better protect the earth and ourselves? I am looking forward to see the answer through the rest of my life.