As far as I know, there are example scripts for imitation learning (IL) with car and with discrete action spaces.
Considering the time it took to make a quadrotor flight with motion planning, it'd helpful to collect data and then use with IL.
Smaller Details:
The lowest level in which quadrotor can be controlled right now is with PWM signals (if not, please feel free to correct me). It would be very useful to have a feature that allows to execute actions (either from human demonstrations or a given policy) and that allows to monitor which PWM signals would have been applied, so that it could be a mapping between the current state and the expected action in terms of PWM signals.
Nature of Request:
Addition
Iam not very familiar with AirSim, but what I have seen it suggests that these changes have to be made at C++ and then passed with the PythonClient API.
Why would this feature be useful?
Research in the fields of Imitation Learning and Reinforcement Learning for low-level control (using AirSim).
What feature are you suggesting?
Overview:
As far as I know, there are example scripts for imitation learning (IL) with car and with discrete action spaces. Considering the time it took to make a quadrotor flight with motion planning, it'd helpful to collect data and then use with IL.
Smaller Details:
The lowest level in which quadrotor can be controlled right now is with PWM signals (if not, please feel free to correct me). It would be very useful to have a feature that allows to execute actions (either from human demonstrations or a given policy) and that allows to monitor which PWM signals would have been applied, so that it could be a mapping between the current state and the expected action in terms of PWM signals.
Nature of Request:
Why would this feature be useful?
Research in the fields of Imitation Learning and Reinforcement Learning for low-level control (using AirSim).