Deep Learning And Probabilistic Methods For Robotic Perception From Streaming Data

Many robots today are confined to operate in relatively simple, controlled environments. One reason for this is that current methods for processing visual data tend to break down when faced with occlusions, viewpoint changes, poor lighting, and other challenging but common situations that occur when robots are placed in the real world. I will show that we can train robots to handle these challenges by modeling the causes behind visual appearance changes. If we model how the world changes over time, we can be robust to the types of transitions that objects often undergo. I show how we can use this idea to improve performance on four different tasks: segmentation, tracking, velocity estimation, and object recognition. Many of the methods in this dissertation are demonstrated in the context of autonomous driving, although they are generally applicable to other robotic applications for dynamic environments. By modeling the causes of appearance variations over time, we can make our methods more robust to a variety of challenging situations that commonly occur in the real world.