data-driven-dynamics
data-driven-dynamics copied to clipboard
ethz-asl
Data Driven Dynamics Modeling for Aerial Vehicles
For some quadrotor UAVs with relatively partial center of gravity, the impact of inertia product should be considered. I_xy,I_xz, I_yz can't just be considered 0(Maybe I misunderstood the code) 
Hello, I have collected some flight data of my real aircraft (quadcopter UAV), can I use the model recognition function directly? Need some other action? I'm sorry to bother you.
**Problem Description** Moment prediction seems to be off by some scaling factor To reproduce: ``` make estimate-model log=resources/quadrotor_model_trajectory.csv ``` - Prediction with QP  - Prediction with Linear Regressor 
As promised last week, I did data gathering flights on 3 different fixed-wing types, each including the same type of maneuvers, and each containing 3 ID phases, marked with `manual_control_setpoint.aux1`:...
- change how flight time calculation is done: use vehicle_land_detected instead of PWM thresholds - OR, what we need for the current work about maneuver-based ID: manually selected start/end points...
Today when playing around with the pipeline I realized that the QPOptimizer failed to estimate the optimal parameters and returned primal infeasible. On the same logs the Linear Regressor worked....
**Problem Description** The current pipeline tries to learn the model from a training dataset generated by uLogs, and evaluates the model on the same dataset.
https://github.com/ethz-asl/data-driven-dynamics/blob/0dca4d349f9c9e338874768b40b7351eace62017/Tools/parametric_model/src/models/rotor_models/rotor_model.py#L164-L165 If the two parameters rotor_position and rotor_axis are collinear the resulting features are all zero. This can cause issues in the regression since the features are all non descriptive.
**Problem Description** It would be useful if we can estimate non dynamics related parameters such as input-current consumption of the motors
Currently the QP optimizer requires all parameters to be either fixed or bounded. We sould add the option for unconstraint parameters as well. @Jaeyoung-Lim what do you thing about just...