SHINE Program 2020

Analyzing Robot Adaptation in New Tasks and Environments

Company/Group: USC, Interactive and Collaborative Autonomous Robotics Systems (ICAROS) lab

Company/Group description: "Enable robots to help people with everyday tasks in human environments"

Role Title: Student Researcher

Role description: Made a program to parse a dataset of recipes to help robots cook

Duration: July 2020 - August 2020

Deliverables: Gave a presentation and poster talk

My Work

The below contain all technial details of my work. The code is available here.

S20-Berkun-Presentation.mp4

Presentation recording

S20-Berkun-Poster.pdf

Poster

What I learned

Skills Exercised/Gained

Python, Anaconda, Jupyter
Confusion matricies

My experience

Take initiative and use your resources

I had a hard time reaching my mentor at the beginning of program. It seemed like everyone else knew what they had to do after the orientation and I was the only one left in the dark. However, after reaching out to both my mentor and the program director, I was able to schedule a Zoom meeting to get started.

I had only 3 meetings with my mentor total, one of them being my final presentation. I didn't think much of it at the time but given my experiences now I would try to schedule more regular meetings with him. I ran into some issues with my code and maybe Slack alone was not the best way to approach that.

Me sliding into my mentor's Slack DMs

My computer trying to run my horribly inefficient code

Code efficiency matters! (For large amounts of data, anyway)

I tried running my code on a small subset of the dataset first. Even that took a number of seconds to run, which made me very concerned about running it on the entire dataset (an few orders of magnitude larger than my test subset).

I sent my mentor my code and asked him if there were any problems, to which he said no. Thus I tried to run my code on the whole dataset. I ended up freezing up the laptop I was using (4 GB RAM) so I asked my dad if I could run it on his computer (16 GM RAM). Using his computer I successfully my program, but it ran for an entire weekend!

Later my brother helped analyze my code with Spyder and found a single instruction was eating up a large portion of the runtime. In the future I would do such an analysis before attempting to parse such a huge amount of data (several GB).

Page updated

Google Sites

Report abuse