Spring 2021 Semester Review
Regarding SEDRo, the plan was to submit a paper on Unity Perception to ICDL 2021. Following that I planned to research more developmental skills for which I could design evaluations. Outside of SEDRo, I wanted to complete the “ Ultimate Guide to Unity Development” course from Udemy, complete reading the Developmental Robotics book. Additionally, I planned to update the documentation for SEDRo side channel, Python API and gaze point detection to SEDRo website. I also wanted to prepare a single document for all my references. In the context of PhD milestones, I wanted to complete my diagnostic.
This semester, I have submitted a paper about evaluation of Unity Perception in artificial agents. I was only able finish the paper on the day of the deadline so I did not have much time to review and revise the paper. There are two reasons I see why I could not finish it earlier. One, my references were all over the place, so it took me a lot of time to find and include them. To remedy this situation, I have now created a single document for all my references with a brief personal note attached to each reference. Two, while writing the paper, I was thinking what the information and logical flow of the paper should be. I think it would help in the future to have a small document dedicated to writing down in paragraphs each concept that I will use for my research. The paper submission did not go through at ICDL 2021. The decision is expected since I myself thought the paper could have been better. The paper could have been more impactful if I provided a run through of the experiments with a baseline model or at least a fake agent. For my next update of the paper, I plan to generate experiment results with a fake agent.
Unity perception is one developmental skill in a plethora of skills infants learn. I researched other skills for which infants can be evaluated for. While search for these skills, I had two main criteria in mind. One, they should be very early level skills. Two, the skills should have a developmental attribute. For example, new born infants does not posses unity perception but the skill develops as the infant grows and shows when the infant is around 4 months of age. I wanted the skills to be early level so that it would be easier to build a baseline model to pass these tests. When I started searching for such skills, there was a lot of information and it was difficult to identify skills that fit the criteria. Then, I thought of Piaget’s stages of developments and remembered that sensorimotor stage is the first stage of development. I narrowed my search for skills developed during this stage of development. Here is a list of papers I went through as part of my research. During this process, I came across a paper, ‘Probing Physics Knowledge Using Tools from Developmental Psychology’, that tests basic physics skills in artificial agents using experiments from developmental psychology. Skills include continuity, solidity, containment, unchangeableness. I thought this is a good place to start since the skills being evaluated fit my criteria. Currently, I am working on replicating experiments from this paper. I made animations for one experiment for each of the skills and I have replicated an experiment for continuity.
As planned, I have documented the building and use of custom side channel for SEDRo. I need to get it reviewed by my peers before publishing it to SEDRo website. I am halfway through documenting gaze point detection. I am yet to work on documenting the Python API for evaluations.
Apart from this, as part of the course I have taken with Dr.Park this semester, I have had the opportunity to read various paper related to my line of research. I have written 12 blogs highlighting the important concepts in these papers. I have also written a suggesting how to read academic paper based on my experience and conversations with colleagues. I’ve also had the pleasure of completing three books this semester namely, ‘How to Grow A Robot’, ‘On Intelligence’ and ‘Society of Mind’. I am half way through the ‘Developmental robotics book’. After reading the book ‘How to Grow a Robot’ I have presented my thoughts. I have started to re-read the ‘Reinforcement Learning’ book by Sutton. I feel that I am in a much better place now when I am reading the book than the last time I read it. I plan to solve the problems in the book as I go. As part of my effort to better my understanding of Developmental Robotics, I enrolled in a course of the same name in Edx. However, I was not able to fully complete the Udemy course on Game Development in Unity.
Regarding my PhD milsestones, I have completed my diagnostic and outlined my thoughts about it here. Outside course and labwork, I have been a part of google’s CSRMP initiative to help students navigate the publication process. The workshop included discussing academic writing, types of conferences and publications, reviewing process, discussing how to ask for collaborations, how to find references for your work etc. It was a productive series of discussions. Additionally, this semester I was CodePath’s tech fellow for teaching Android App development to 25 UTA students.
I feel that I am more focused this semester and feel motivated to work and excel. My teammates are helpful and discussions, especially during weekly presentations, are effective. I feel connected with the work I am doing and I believe I can make good contribution to SEDRo
In the future, I want to work in a research oriented role in the industry. I want to look out for roles that involve simulating human environments and aspects of human behaviour. My focus will be at the intersection of developmental psychology + evaluation of artificial agents + Unity platform.
Challenge this semester - All my previous work on replicating experiments was gone as Unity was not functioning properly for me. I had to recreate all my work. I plan to make multiple copies of my work. Documenting my work helped me recreate my work with much less effort
I think connecting virtually has provided more flexibility for me and gives me more time to work. I think it would be great if we communicated more amongst ourselves (labmates) and kept each other update to date on our work. More brainstorming sessions between us would be helpful. I really appreciate that Dr.Park is available to us and invests time with us to improve our research and research careers.
I am trying to follow the lab culture to the best of my abilities. I collaborate with my lab mates whenever possible. I try to make my work reproducible. I am earnest in my work. I am working on making my presentations better so that they are easy to understand. I have the drive to learn and excel.
In the coming semester, I plan to complete replicating experiments from the probing physics paper. I would also like to generate results on the Unity Perception experiments using a fake agent and update my paper with the results. While working on the probing physics paper, I would be focusing on the following questions.
How to project the examples? - screen or physically
How to perform and evaluate habituation
Look at the VOE method proposed in the paper in detail
How many examples to show in the training phase
I have made a plan to strengthen by understanding of developmental robotics and psychology. I plan to stick to my plan and further my understanding. Additionally, I want to complete the game development and animation in Unity courses from Udemy.