Gazebo F1 Follow line dqn & qlearning comparison (months 29, 30, 31, 32)

less than 1 minute read

Finished cartpole!!

Next goal

Next challenge is to revisit the F1 follow line problem in RL-Studio and accomplish a performance analysis of already implemented algorithms making use of Behavior metrics.

After that, we will be ready to step out into carla, which will be the tool used for my thesis.

That said, the following is the progress got from the two first months of work in F1 Follow line:

Progress

  • qlearning running and learning with simplified perception and discrete actions
  • dqn running and learning (some adaptations were needed to run) with simplified perception and discrete actions
  • qlearning v1.2 migrated to Behavior metrics
  • dqn v1.2 migrated to Behavior metrics
  • Behavior metrics script to automatically compare several algorithms in several circuits in Gazebo fixed
  • dqn and qlearning trained in RLStudio brains compared and plotted in Behavior metrics

For more details, see the qlearning - dqn gazebo f1 follow line comparison blog and the follow-line dqn refinement and [the follow-line dqn comparison](https://roboticslaburjc.github.io/2020-phd-ruben-lucas/projects/2023-05-29-F1_follow_line_dqn_comparisson/

Blockers:

  • ddpg is not running in rlStudio and therefore barely learning anything

Next steps:

  • make ddpg work

You May Also Enjoy