Wilbur is a robotics project, a personal captstone, and an experiment in embodied cognition - the notion that only an agent interacting with the demanding complexity of the real world (or a digital representation), can develop true general intelligence.
To this end, in this repository I will explore the interplay of computer vision (CV), natural language processing (NLP), and reinforcement learning (RL) in progressively complex phases that complement Wilbur's overall development. Because I'm a huge fan of Marvel, the project is naturally organized into "Marks", an unabashed homage to Tony Stark's various Iron Man prototypes:
Mark One was principally built to study objection detection and segmentation, but also uses RL techniques to track objects in its viewport. In the mark_one/models folder, you'll find the following preliminary models:
resnet_sandbox: a Jupyter Notebook containing my implementations of VGG and ResNetunet_sandbox: a Jupyter Notebook containing my implementation of U-Net.retinanet_sandbox: a Jupyter Notebook containing my implementations of FPN and RetinaNet
Coming soon.
Coming soon.