- Research Internship, Dubai Future Labs – Dubai, United Arab Emirates (Dec 2021 – Jan 2022)
During my Winter break after my first semester at CMU, I came to the Dubai Future Labs, where I had the opportunity to work on a couple of projects.
- The first was to integrate the Bitcraze Crazyflie drone with ROS for swarm formations. I was quite impressed with the robustness of these drones as well the extensive documentation and ROS support that is already available, making it a great choice for researchers starting out with experimenting with drones. These drones are also quite inexpensive starting at only $225. In addition to this, it also has a range of extension decks that make it really easy to customize the drone according to your preferences. My favorite was the AI Deck which uses a camera and allows you to perform visual SLAM or deep learning-based tasks all on-board! Unfortunately, I did not have the time to explore all its capabilities during my month there, but definitely will in the future. I used the Crazyswarm ROS package which is developed and maintained by USC’s ACT lab to achieve position control using Optitrack system for localization. The optitrack system broadcasts the locations of rigid bodies created by fiducials over VRPN. The setup and configuration was extensive I had to figure out nuances not mentioned explicitly in their documentation. During this time, I collaborated with Amir Abdou during this process, and we have documented the process we followed here.
- We were able to achieve position controlled drone formations involving four drones, and also embedded the inbuilt API that performs collision avoidance between the drones in the resulting swarm.
- For my second project, I am collaborating with Ahmad Al Attar and assisting him with his research on model free control of robot arms. The idea is to design data-driven controllers that can control robot arms without any prior knowledge of its model, and purely by doing some explorations and exploiting that information gained. The work we are doing is based on the earlier work by Dr. Peter Kormuchev , Senior Lecturer at Imperial College London, which you can learn more about by reading the paper that he and his colleugues presented at ICRA 2015.
2. Autonomy Engineer Internship, Dubai Future Labs – Dubai, United Arab Emirates (Dec 2021 – Jan 2022)
- Before the commencement of my Master’s education at CMU, I worked as an autonomy engineer intern at Dubai Future Labs. Leveraging my prior experience in ROS, I was able to architect and integrate the ROS navigation stack to a customized indoor warehouse automation robot, from mapping, sensor fusion for localization, and autonomous navigation. Since the robot housed a combination of cameras, LiDARs, and IMUs, I gained valuable experience in understanding the merits and challenges involved in calibrating these sensors both intrinsically and extrinsically. I also developed a simulation environment using Gazebo for testing out various robot functionalities before deployment on the real robot.
- During this process, I was also constantly interacting with the hardware team to simulate several design iterations. I also designed and implemented a human-follower functionality that involved the complete detection, tracking, and control pipeline. The detection was carried out using an off-the-shelf ROS implementation of YOLO-v3 module, which was deployed on a Jetson Xavier computer. I then integrated the detection module into a UKF-based tracking module, which was robust to occlusions. A customized python script was then used to integrate and control the various sub-components to facilitate human following.
- Since I was working with a system in which various components were deployed on different computers connected over a network, I was able to appreciate the importance of ensuring synchronized clock timings between the systems and the effects of communication latency on the overall functioning. Moreover, having worked with a mature software stack with multiple collaborators, I gained valuable lessons on the development of unit tests, system integration tests, and ensuring that the code I write is as per the development standards.
3. Robotics Research Internship, Indian Institute of Science – Bengaluru, India (Nov 2020 – Jul 2021)
- I also undertook research internships at India’s best research university, The Indian Institute of Science, both during and shortly after my undergraduate education. My latest stint involved the development of a controls pipeline for a lightweight table organization robot arm which involved using a robotic arm to rearrange various objects on a table to a predefined configuration. I had a wonderful collaborating with my colleagues Gaurav Sethia and Siddharth Ghodasara.
- This project involved various challenges such as grasping in clutter and rearranging into complex configurations that involved object stacking. My team and I worked on this project as a part of the Open Cloud Robotics Table Organization Challenge , the world’s first Cloud Robotics challenge, where participants could test their solutions remotely over the cloud on a robot arm after having developed their solutions in simulation. It was conducted as a part of IROS 2020 and our submission made it among the top 20 participants. We continued to work on this problem statement after the competition, and a publication is currently under review at ICRA 2021.
- These experiences of working in academia taught me the rigor of scientific research and the amount of time and effort that goes into developing novel work. I learned to reason about a first-principles approach that helped me greatly in thinking about solutions from a different perspective. I also learned the importance of writing modular code that can easily facilitate ablation studies to compare the efficacy of our work with the state-of-the-art. The video below is a demonstration of our solution for table organization on a YASKAWA Moto Mini robot arm.
4. Robotics Research Internship, Indian Institute of Science Bengaluru, India (Jan 2020 – Jun 2020)
- During the final year of my undergraduate education, I had the opportunity to work under Prof. Abhra Roy Choudhary, at the Robotics Innovations Lab (RIL) at the Indian Institute of Science, Bengaluru. During the first phase of this internship, I conducted a research study on parameters affecting the effectiveness of human-robot interaction at Open Day, IISc. I collaborated with Aiswarya M during this phase of the research study, and the setup is as summarized in the video below:
- However, after the onset of COVID-19, operations in the lab had been transformed to remote operation. During this time, since we did not have access to any real robots, the remainder of my internship was carried out in simulation. I implemented computer-vision based automated fault detection and segregation in spur gear assembly plants using a combination of ResNet-50 and classical techniques. This pipeline was validated in simulation using CoppeliaSim and MATLAB. The video below explains the process adopted.
5. Robotics Internship, Indian Institute of Science – Bengaluru, India (Jan 2020 – Jun 2020)
- This was a month long internship at the Stochlab at Robert Bosch Center for Cyber Physical Systems, Indian Institute of Science. I had the opportunity to contribute to the development of libraries written using C and Python for control of KONDO servo motors in quadruped robot ‘Stoch’.
6. Internship/Part-time, Li2 Innovations – Bengaluru, India (Sept 2016 – Oct 2019)
- I worked with Li2 Innovations initially as an intern and then collaborated part-time during teh course of my undergraduate education. I contributed to the upgrade of the “BRAMHA-3” 3d printer to facilitate IR-sensor-based auto-bed leveling.
- Apart from this I also developed learning modules and served as a workshop trainer for 10000+ high school and undergraduate students from schools and universities across India. I absolutely love to teach and I continue to teach Robotics and AI whenever possible.
- Most of my learning modules have now been converted to full-fledged MOOCs on a platform called ThingQbator, supported by the Nasscom Foundation.