Shravan Kumar Gulvadi
I am a graduate student in Mechanical Engineering with a specialization in Robotic and Control Systems from Carnegie Mellon University. I am interested in making robots work under uncertainty and clutter of real world environments and in designing the necessary robotic software and hardware to achieve this. I have 4.5 years of professional experience in Product Development and have designed automation solutions and new products that have been successfully deployed in extreme operational environments.
Mainspring Energy (January 2023-Current)
Designed hierarchical state machines for subsystems to guide system behavior
Implemented state machine and controller designs using C++
Gained experience in writing unit tests using the Catch2 framework
Graduate Research Assistant
Artificial Intelligence for Robot Coordination at Scale (ARCS Lab) (August 2022-Current)
Working under Prof. Jiaoyang Li on developing Robust Multi-Robot Motion Planning Algorithms.
Co-designed a motion planner that augments advanced robust control methods with multi agent path finding techniques. This algorithm allows the planner to reason about uncertainties associated with execution while planning.
Robotics Research Intern
Voaige (May - August 2022)
Developed the Eye-Hand calibration software for robotic arms, implemented an auto calibration algorithm from scratch.
Co-designed the ROS architecture for adaptive general purpose robot arm control and wrote robot services for a novel vision enabled pick and place application on a 6 DOF UFactory X-Arm.
Graduate Research Assistant
Computational Engineering and Robotics Lab (January-May 2022)
Collaborated with a team of researchers on augmenting traditional control methods and reinforcement learning to come up with policies to stabilize aerial drones and underwater vehicles in high turbulence environments.
Implemented a wind estimation model and trained the Reinforcement Learning agent to perform gain scheduling of PIDs depending on the current sensed state of the drone.
Deputy Engineer - Product Development Group
Bharat Electronics Limited (September 2016-May 2021)
Handled multiple projects from ideation to design, manufacturing, integration, testing-validation and customer offering. Experienced working in cross-functional teams and as individual contributor.
Designed multiple Automation solutions for defense platforms. These are mission critical systems that had to function in high shock, extreme weather conditions.
Designed multiple products as per stringent MIL and Aerospace standards.
Mentored 8 engineering trainees and 4 diploma trainees by providing technical training.
Won ‘Certificate of Recognition’ for innovative design and ‘Suggestion Award’ for suggesting and implementing key process changes that improved the process and opened new business avenues for the company.
Master of Science in Mechanical Engineering - Advanced Study
Carnegie Mellon University
Concentration in Robotic and Control Systems
Current Grade: 4.0/4.0
Course Work: Modern Control, Statistical Techniques in Robotics (Machine learning for Robotics), Computer Vision for Engineers, Multivariable Control (Robust Control), Bio-inspired Robot Design and Experimentation, Simultaneous Localization and Mapping, Engineering Computation, Engineering Optimization (Non-Linear Optimization), Advanced Control System Integration, Path-planning and Decision making for Robots
Bachelor of Engineering in Mechanical Engineering
The National Institute of Engineering
Implemented trajectory planning algorithm and the tracking controller to perform multiple flips on a crazyflie drone
Designed the controller in simulink and tested it in the Gym-pybullet-drones simulation environment and successfully ported the planner and controller to the hardware
The Goal of the project is to enable autonomous navigation of an autonomous UAV in a forest environment to allow UAV and UGV robot teams to identify and eliminate potential triggers of forest fires
Implemented Simultaneous localization and mapping algorithm that utilizes semantic information to enable autonomous navigation of drones
Designed controllers to control Tesla model 3 in a simulation to successfully complete a lap around the CMU campus.
Involves linearizing the vehicle dynamics model, designing a PID, full state controller and an optimal controller (LQR).
A* algorithm implementation for planning and implementation of Extended Kalman Filter for simultaneous localization and mapping.
While extensive research allows most animal behavior patterns to be understood, one phenomenon still leaves researchers puzzled: the tendency of lizards to switch from a quadrupedal stance to a bipedal stance while running. The purpose of this project is to investigate any variations in energy consumption that some lizards may experience during motion when they actively enter a bipedal gait before natural rearing can occur. Prior research suggests that this active transition to a bipedal gait may occur in certain lizards, but offers no insight as to what advantages this behavior may offer over rearing naturally. The team hypothesized that lizards could achieve lower energy consumption by initiating active rearing during motion. Through experimentation on a lizard-inspired robot capable of modeling both passive and active rearing, the team hoped to determine the amount of time spent passively rearing after which our system would gain an energetic advantage by actively rearing.
Current research work is focused on augmenting Machine Learning methods with traditional control methodologies to develop more robust and better performing controllers. Work done as a part of a research team in CERLAB at CMU.
Implemented Dense SLAM with point based fusion for 3D localization and mapping. Implemented Projective Iterative Closest Point algorithm for Visual Odometry and glued it with point based fusion for Simultaneous Localization and Mapping.
Implemented Particle Filter/Monte Carlo Localization to localize a lost robot in an indoor environment. Implemented robot motion model, sensor model and ray tracing algorithm for successful 2D localization of the robot.
Worked for an early stage startup ‘Voaige’ to come up with an Online eye-hand calibration methodology suitable for their application and wrote the software for the same
The goal was to take the image from the camera as input and to map the coordinates of the objects in the image to the real world with respect to the robotic arm
Worked on literature survey studying latest methodologies for eye hand calibration, image formation in cameras, algorithm design,testing and deployment on the robot
Designed a Model Reference Adaptive Controller to stabilize a DJI Mavic drone.
Simulated a case of single rotor malfunction and demonstrated the capability of the controller to stabilize the system in cases leading to uptown 70% of thrust loss in the rotor.
Compared the performance of the adaptive controller with the LQR and demonstrated its advantage.
Implemented Model Predictive Controller (MPC) to balance the cartpole in swing up position.
The MPC has two components, the first tries to learn the environment dynamics and the other plans the policy (sequential actions) based on the current state.
Implemented Iterative Sampling-Cross Entropy method for the planner, the algorithm iteratively optimizes a set of Gaussian distributions that model a distribution over each action in a trajectory.
Python, C++,ROS, Matlab, Simulink, Tensorflow, Pytorch, Embedded C, Open CV, Linux, Git
CAD & Simulation
Solidworks, Webots, Ansys, Autodesk CFD
3D printers, Laser Cutters, Turning & Milling, CNC, Soldering, Hand tools