Computational Design of Bio-Inspired Fliers
Guided by Professor Tao Du
RL-Trained Control Policy for a Rigid Butterfly
We develop a reinforcement learning based control policy for a rigid butterfly model. The policy learns to generate stable, efficient flapping motions mimicking the real world butterfly.
XPBD Wing Simulation
We employ an Extended Position-Based Dynamics (XPBD) framework to simulate deformable wings, using a coupled shell and rod model. The solid simulator is further coupled with a lattice Boltzmann method (LBM) fluid solver.
Simulation Performance
Influence of Wing–Wing Vortex Interaction
Comparing to simplified fluid models, we want to address the importance of using a higher precision fluid solver. A key limitation of simplified models is that they cannot capture the interaction between multiple wings. We let two wings flap with their tips close together. Our results show that when the right wing flaps in phase with the left wing, it increases the force experienced by the left wing. When the right wing remains stationary, the force on the left wing decreases by about \(10\%\). A further \(10\%\) reduction occurs when the right wing flaps at a phase difference of \(\pi\). These observations highlight the complex aerodynamic coupling that occurs in groups of insects flying in proximity.The image below shows the influence of the vortices generated by the two wings.
Hardware Design
We also design hardware to validate our results. Thanks for the help of my collaborator Jiaxi Mei.
Sim2Real
Our hardware test characterizes the remaining sim-to-real gap. To evaluate both the accuracy of our hardware-based force measurement system and our simulator, we design three experiments and compare the measured forces against simulated predictions. The motion of the flapping wing in these experiments is shown below.
