The Weekly Carca, a weekly forum.
A social networking platform for intellectual, opinionated, and inquisitive people, where they can interact with a diverse set of people through audio-based group discussions. The Weekly Carca was part of Carca as a pilot forum for discussions every week on various pressing topics, and for people to learn about them from different experts in the fields.
February 2021
The negative impacts of forest fires are increasing dramatically, driven by the effects of climate change and other factors. Robotic systems are one method to improve the feasibility of prevention and mitigation efforts. In this work, we propose an unmanned aerial vehicle (UAV) that can map a region, so an unmanned ground vehicle (UGV) can autonomously clear fuel to prevent the spread of fire. We developed a multi-sensor payload consisting of cameras, LiDAR, and GPS with onboard processing. We also implemented a SLAM system to understand the 3D structure of the environment, a semantics system to identify fuel and other features in the environment, and a traversability system which predicts which region a UGV can navigate. This approach provides a 3D map of the environment and georegistered maps describing the locations of fuel and traversable regions. We validated our system with preliminary field trials and showed that this is a promising approach.
Safe planning and parking demonstrated in non-holonomic robot in CMU RI Basement.
Autonomous, non-holonomic wheeled robots are being increasingly used in unstructured environments for operations like exploration, search and rescue, etc. Many situations may require robots to park themselves during navigation and operation in these types of environments. These situations can arise due to various reasons such as sensor failure, unfavorable operating conditions, etc., but specifying a safe parking spot and navigating to it safely can be difficult in unstructured environments.
Together with my team I developed a planner for non-holonomic wheeled robots to reach any given goal position and orientation safely in unstructured environments with multiple obstacles. Also developed a novel solution to efficiently find a safe parking location for the robot in its surrounding environment. This allows the vehicle to either safely stop when faced with adversarial scenarios or allow other robots in its team to navigate through the environment.
December 2022
Eye-in-Hand calibration data flow.
Autonomous drones are being used in multiple industries, such as payload delivery and search and rescue. A common requirement for drones in these situations is to carry as much material as possible in each flight, which can be achieved by combining the payload capacity of multiple drones. I worked with my team to realize this as a dual connected quad-copter system and designed a system that is capable of performing aerial maneuvers and tasks. We developed custom hardware and software components, including the design and fabrication of the link that connects them. The dynamics of the combined system were defined and used to create a control system for safe hovering and trajectory following. We demonstrated the potential of dual connected quad-copter systems as a powerful and versatile tool for a wide range of applications.
December 2022
Video colorization of India 4K using CVC.
Colorization of images using deep learning is a problem that has been tackled in multiple ways over the years. Video colorization has the added complexity of maintaining temporal consistency across multiple frames. Although this problem has gained momentum in recent times, there has been much less research effort focused on video colorization compared to image colorization. I worked on implementing different baseline architectures of CNNs, visual attention, generative models, and experimenting with different loss functions to obtain better fully-automatic video colorization results.
July 2022
Eye-in-Hand calibration data flow.
As part of a team, we aimed to calibrate the positioning of a robotic arm with a camera mounted on the end-effector using a SLAM-based method. Specifically, we used the Extended Kalman Filter (EKF) to achieve this goal. To approximate the camera's extrinsic parameters, we utilized its intrinsic parameters and a set of feature markers in a workspace. We also deployed an EKF-based measurement model to effectively localize the camera and compute the camera-to-end-effector transformation. During the project, we tested this approach on a UR5 manipulator with an Intel RealSense D435i depth camera mounted on the end-effector and successfully validated our results.
May 2022
KickFlip final prototype.
The KickFlip - The Cooking Robot to automate frying and grilling within a home kitchen and a small-scale restaurant. Potential alternative to more expensive options like Flippy, Popcake and Cinder Grill in the market. We designed the prototype of KickFlip to save time and effort in the kitchens through increased autonomy during the cooking process through food monitoring and interaction.
December 2021
Aurecle pipeline video sample estimation results.
AuReClE can be retrofitted into existing logistic vehicles with a perception system and a heads-up display for assisting the drivers. Using a vision pipeline (image processing and neural network), our system will estimate the clearance in the upcoming structure, and based on the vehicle dimensions, it would give a go-no-go warning through a heads-up display to the driver to prevent any injuries and damages.
December 2021
One of the tested solar cooker designs (SK-14).
Solar Cookers invariably depend on various factors that affect their performance and efficiency. The amount of irradiation available being one of the prime affecters, the wind speed as well as the cooker design affects its performance situationally. Conducted research and testing on different types of solar cooker designs and compared their performance parameters and efficiency based on ASAE standards.
December 2019
Pratt Truss structure stress and strain analysis
Built a MATLAB GUI to perform static structural analysis on different truss structure designs such as Gambrel, Howe, K-Truss, etc. Any custom truss structure could be drawn in the internface and its stress vectors were generated based on force applied at a certain point on the truss using FEM.
May 2019
CAD Model of the fabricated walking mechanism.
Designed and fabricated a 4-legged walker primarily to learn different production. But, also to design a locomotion mechanism that can be used to control easily with just a single actuator and can be implemented in field robots. The walker's mechanism was based on the Jansen’s linkage as it has SDOF and operates smoothly in various conditions; Also, it looks really cool in operation!
December 2018
Model of the proposed airborne wind turbine.
The future of high altitude wind energy is vast as the winds are stronger and more stable at high altitudes The Airborne Wind Energy System (AWES) was a proposed prototype system to harness power from high-altitude winds and produce energy more efficiently compared to the traditional ground windmills. At the same time solve the various drawbacks posed by the traditional ground windmills, such as noise, ecological impact, high cost, and immobility.
April 2018
One of the levels in the game.
Move it! is inspired from the “The World's Hardest Game.” The game was developed as a C++ GUI application with different functionalities and various pre-made levels. The game also incorporated a custom game level making tool which allowed the players to make and play their own levels.
March 2016