I am interested in intelligent systems that blend perception, natural language processing, and machine learning to guide agents in the real world. In more than 20 years at the Robotics Institute at Carnegie Mellon University, I have worked with many intelligent systems, including autonomous vehicles operating in structured and unstructured environments, advanced concepts in vehicle safety, and machine vision for manufacturing, among others.

Deep learning presents the opportunity to reevaluate how we design complex systems that are frequently composed of a group of modules. Often, these modules are engineered to individual performance metrics that may only partially support the overall task. Combining previously discrete modules into a single deep network and loss metric can improve performance while simultaneously simplifying the design and increasing flexibility.

My recent doctoral thesis describes a robot navigation system that accepts text commands and aerial maps to generate paths for a ground vehicle. This work fuses elements of scene labeling, natural language processing, symbol grounding, and path planning. The architecture uses imitation learning to mimic human behavior, simultaneously optimizing the component subtasks to learn the mapping from input data to paths without any intermediate steps or strong modeling assumptions.

Curriculum Vitae

CV 

Contact

Email: arnejsuppe@gmail.com