**Current Projects:
Mastery in Minutes: Deep Reinforcement Learning for Robust and Sample Efficient Large Scale Robotics Platforms
Recent advances in equivariant learning, efficient training, and unsupervised representation learning have accelerated deep reinforcement learning of robot manipulation policies (DRL-RMPs) in real robots. DRL-RMPs today can train within an hour. However, learning robust DRL- RMPs from a single or a few demonstrations remains an open problem. Such policies also remain brittle and have limited generalization. This work seeks to learn robust generalizable DRL-RMPs from less than 10 demonstrations and train within 20 minutes. We seek to innovate theories and algorithms at the intersection of representation learning, group theory, and data-efficient RL.
Research for Undergraduate Students
My research delves into how robots can learn to manipulate and interact with the world in more intelligent ways. I research robust deep-learning algorithms that can be applied to various scenarios, including services, manufacturing, and human-robot interactions. We are interested in developing algorithms that can learn faster, better, and from less experience.
Research begins with a passion for solving problems, a curiosity to learn, and the determination to develop the skills to get there. In pursuing research early on, you will nurture a spirit of research-inquiry and develop strong skills in programming, probability and inference, artificial intelligence ( machine learning (supervised, unsupervised), deep learning (reinforcement learning, imitation learning) and more.
You will be a full-time user of Linux, ROS, Torch and/or Pytorch, and simulators. You will use editors like vim, emacs, and usually write in LaTex in Overleaf.
We publish our research in top international conferences and journals, as seen in these robotics Google Scholar metrics. Interested students should be aware of the latest scientific contributions in the fields related to the lab. I expect students to build their ‘idea stack’. That is, 2-3 research hypotheses they would like to explore. These ideas will guide you amid waves of information, papers, and new paradigms.
Students with a implemented projects are attractive if you have them. Interesting publication venues in the area of robotics are listed here. In joining my team, you will become acquainted with the literature in the fields of robotics relevant to our group; (2) build a technical background in math, probability, and AI that supports independent and innovative research (3) master experimental tools and methods of empirical analysis; and (4) the development of communication skills for the clear and effective presentation of your results orally and in written form.
Contact:
Interested candidates should also send their inquiries to Dr. Juan Rojas at juan [dot] rojas [@] lipscomb [dot] edu
Would you like to do research together?
I welcome undergraduate and high-school students as well as collaboration with any graduate student or program. Over the years, I have worked with post-doctoral fellows, PhD students, Master’s students, and undergraduate students in publishing high-quality papers at top journals and conferences.
General Advice
In this section, I I would like to share general advice that I have found helpful.
On Research…
Many of these come from Bill Freeman from CSAIL MIT
How to do Graduate-level Research:
Some Advice by Dr. Bhaskar Krishnamachari
How to do Research
by Bill Freeman from CSAIL MIT
How to have a Successful Research Career
by Bill Freeman from CSAIL MIT
On Publication…
How to write a good CVPR Submission
by Bill Freeman from CSAIL MIT
On Machine Learning…
Advice on Machine Learning Projects
by Tim Rocktaschel
On Careers in Academia…
Check this study based on surveys of post-docs