Description
Aerobat is a bio-inspired flapping-wing robot that explores agile flight under highly coupled, nonlinear dynamics. I lead the perception, state estimation, and simulation pipeline, enabling onboard localization and closed-loop control without external motion capture.
Project Details
I designed the entire perception and state estimation stack, including writing multithreaded C++ drivers to reliably synchronize IMU and camera data. This enabled robust visual-inertial odometry on a challenging, highly dynamic flapping platform. I also formulated and implemented a custom EKF (Extended Kalman Filter) that explicitly models elastic coupling between the flapping robot body and the control frame, producing high-rate, physically consistent state estimates essential for closed-loop control.
In parallel, I developed a MuJoCo-based simulation that realistically captures flapping dynamics, aerodynamics, and compliant coupling, enabling thorough simulation-in-the-loop testing prior to hardware deployment.
Aerobat operates far outside rigid-body assumptions—compliance, vibration, and delay are major challenges, not exceptions. The methods I developed here are robust to these effects, advancing state estimation and control for soft or hybrid robotic systems.