Duration: May 2025 - Present
- Proposed a hierarchical control framework for agile trajectory tracking on the Multi-Modal Mobility Morphobot (M4) (6 kg), a morphing aerial robot capable of transitioning between ground and aerial configurations, under the guidance of Prof. Alireza Ramezani.
- Developed the NMPC trajectory optimizer using nonlinear optimization (CasADi/IPOPT), a QP-based real-time thrust allocator, and a thrust-to-PWM mapping pipeline for hardware deployment on a conventional quadrotor platform.
- Demonstrated in simulation that actively reconfiguring the robot's appendages across aerodynamic regimes to utilize posture manipulation and thrust vectoring outperforms fixed-geometry baselines in tracking error and thrust redistribution, enabling high-speed sharp turns of up to 120°.
- Validated QP-based real-time thrust allocation under active posture manipulation, demonstrating stable hover throughout randomized joint configuration changes.
- Performed bench-top motor-ESC characterization via frequency-domain system identification on a VOXL + ESC stack, estimating system parameters (gain K, time constant τ, delay L) from step and sinusoidal excitation using Nonlinear Least Squares and Bode analysis; validated by inverting the model into a force-to-PWM mapping.
Project Duration: January 2023 - March 2024
- Worked as a research intern at the Multi-Robot Autonomy Lab at IISER Bhopal under the guidance of Dr. P. B. Sujit and Dr. Manoj Kumar Tripathi.
- Co-developed CFDMPC, a novel framework integrating PINN-based wind field estimation with NMPC (CasADi/IPOPT) for wind-aware UAV path planning in cluttered environments, achieving 100% collision-free navigation across all obstacle configurations while the constant-wind baseline crashed in every multi-obstacle scenario.
- Trained a data-free PINN using DeepXDE (TensorFlow backend) to solve steady-state RANS equations without precomputed simulation data, enabling millisecond-level wind field inference at arbitrary spatial locations compared to 5 wall-clock hours for traditional CFD solvers.
- Validated robustness under randomized inlet velocities and directions; demonstrated PINN-predicted wind fields outperform panel-method estimates that misguide the planner into turbulent high-shear gap regions between obstacles, reducing control effort ~9% vs. a full CFD-driven planner.
Project Duration: June 2022 - July 2022
- Developed Prota: The ROS Bot as part of the e-Yantra Summer Internship, a low-cost open-source educational autonomous ground vehicle designed to teach ROS, SLAM, and navigation from the ground up.
- Integrated LiDAR, IMU, optical encoders, depth camera, and proximity sensors on a Raspberry Pi; calibrated and synchronized all sensor streams for reliable odometry and localization in a GPS-denied environment.
- Deployed GMapping and Hector SLAM, used AMCL for localization and ROS move_base for autonomous navigation; validated the full navigation stack first in Gazebo/RViz simulation and then on physical hardware.