A CARLA-based study finds that masking invalid steering actions and using relative steering adjustments can accelerate PPO training convergence, but the gains over full action spaces are mixed and unverified.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Action Space Reduction Strategies for Reinforcement Learning in Autonomous Driving
A CARLA-based study finds that masking invalid steering actions and using relative steering adjustments can accelerate PPO training convergence, but the gains over full action spaces are mixed and unverified.