SHINE 2020 -- Professor Nikolaidis ICAROS lab
To advance robots, engineers strive to make robots more adaptable to the kinds of real-life environments humans often encounter. In a kitchen environment, for example, a robot may not have access to all the tools it would expect to use, and it may be commanded to carry out tasks it has never seen implemented before. Our research aims to develop an algorithm to choose the best available tool for a given task (ex: "cut banana"). My role in all of this is to generate training data from online datasets and human feedback so the robot can learn what tools are best for certain types of actions.
Our research will help develop robots to assist humans in a cooking environment, where certain tools are not available and where commands without mention of a tool (ex: "cut the banana") are often used by humans. Eventually, we hope our research can be expanded to help other robots generalize learned manipulation action plans to new tasks and environments.
In this repo, I have attached all the code I wrote during SHINE 2020 and all of the code's inputs and outputs, with the exception of layer1.json, which was too large to upload. You can download layer1.json from the website provided in the recipes_dataset_link.txt file.
All files are my original work. Each commit description serves as a description of the file's function and purpose.
For a general overview of my project check out my padlet.
I would like to thank Professor Nikolaidis, my PhD mentor Hejia Zhang, and my SHINE partner Nikitas Klapsis for helping me in my research journey. I also want to thank my center mentor, Ashley, and Dr. Mills for checking in with me and making sure I wasn't confused about SHINE. Lastly, I would like to thank the rest of the SHINE team and participants -- without you, none of this would have been possible.