This project is advised by Dr. Alan Fern. In this particular task, I have developed the maneuver to pick a ball that is surrounded by a lava region (untraversable) by using a tool to push it in the traversable region and pick it up. The task is challenging because the robot can only see from its perspective view analogous to a human being and the room map is unknown. We stood a 2nd place while competing with CMU, UCB, MIT and IBM.
Developed a multimodal neural network consisting of CNN and transformers, a novel approach to classifying audio data and also taking an advantage of lyrics to improve the accuracy. Our results showed that our model had 34% improvement as compared to the prior multimodal and uni-model approaches.
Cassie is a bipedal robot that smashed the Guinness world record for fastest 100m run. Currently, the command is provided by human to traverse in the environment which mean Cassie is blind. I am working with making it autonomous by putting camera module and building robust heuristic.
Developed a model-free self-balancing maneuver through in reinforcement learning framework. This approach is beats the prior work on self-balancing that used a PID controller or model-based controller because oftentimes it is hard to understand the system and it lacks the robustness. With learning approach, I was able to show it is relatively easier to design regardless the complexity of model and yet it is robust.
An android application that records audio / video, captures image even when the device screen is off. Start recording anytime with volume button and capture anything in disguise. It has 200K+ downloads, 4.1 ratings with over 500 daily active users.
I have found a major bug and proposed a solution to the HyperTrack.. HyperTrack is toolset contains iOS, and Android SDKs which generate live location data, that is then consumed through webhooks and APIs.
I am doing master's in Oregon State University. I am interested in Artificial Intelligence, particularly in Reinforcement learning and Robotics. I am stunned by how they do near-human job and sometime better.
Building viable products
It is always better to bring your ideas to the market. I would love to experiment my learning and build something that is useful in day-to-day life.