search

lis-epfl / apg_trajectory_tracking

Training efficient drone controllers with Analytic Policy Gradient
115 stars 15 forks source link

A question of quad_dynamic_simple position step. #4

Closed HenryHuYu closed 9 months ago

HenryHuYu commented 1 year ago

Hi, Great work you've done! When I read the code in quad_dynamic_simple.py, I found px_new = px + 0.5 * dt * dt * acc_x + 0.5 * dt * vx py_new = py + 0.5 * dt * dt * acc_y + 0.5 * dt * vy pz_new = pz + 0.5 * dt * dt * acc_z + 0.5 * dt * vz

I am really confused about the 0.5 before dt * v? could you plz give a hint about it?

also could you plz explain what is a "torch_kinv_ang_vel_tau" varieble? Thank you very much

NinaWie commented 9 months ago

Hi, thanks for your interest and very sorry for the late reply! As said in our paper, our environment is based on the Python implementation of a gym environment for quadrotors available here. We use the integration from that implementation, see here. Then, we made the quadrotor model more complex and aligned it with the model implemented in the flightmare simulator. In Flightmare, the "kinv_ang_vel_tau" is defined here and is used as a constant in computing the body torque. I'm closing this for now, feel free to reopen.