search

learn-to-race / l2r

Open-source reinforcement learning environment for autonomous racing — featured as a conference paper at ICCV 2021 and as the official challenge tracks at both SL4AD@ICML2022 and AI4AD@IJCAI2022. These are the L2R core libraries.
https://learn-to-race.org
GNU General Public License v2.0
146 stars 15 forks source link

Recording human demo runs #10

Open trancenoid opened 2 years ago

trancenoid commented 2 years ago

Is there any way to play (human input using keyboard) the game and record the RGB, Segmentation, Actions and Rewards to augment the demonstration dataset ?

trancenoid commented 2 years ago

Found a way : Add the sensors to the Simulator Config, call env.make() call env.record_manually(output_dir)

trancenoid commented 2 years ago

I noticed some problem with this method. I am able to get the Observation (All segmentation, RGB views along with multimodal values), but there is no way to get the actions that I inputted to the simulator while recording my run. I looked into the code but because the _observe function pulls only the sensor reading, inputs given directly to the simulator using the keyboard is not captured. Is there a way to record actions too in record_manually function ?