Curriculum goal masking for continuous deep reinforcement learning
This work addresses the challenge of inefficient goal sampling in continuous deep RL for robotic manipulation, offering an incremental improvement over existing methods.
The paper tackled the problem of optimizing goal sampling in deep reinforcement learning by introducing a curriculum goal masking method that estimates goal difficulty, showing that focusing on medium-difficulty goals works best for DDPG and hard goals for DDPG+HER, with significant performance improvements in robotic object manipulation tasks.
Deep reinforcement learning has recently gained a focus on problems where policy or value functions are independent of goals. Evidence exists that the sampling of goals has a strong effect on the learning performance, but there is a lack of general mechanisms that focus on optimizing the goal sampling process. In this work, we present a simple and general goal masking method that also allows us to estimate a goal's difficulty level and thus realize a curriculum learning approach for deep RL. Our results indicate that focusing on goals with a medium difficulty level is appropriate for deep deterministic policy gradient (DDPG) methods, while an "aim for the stars and reach the moon-strategy", where hard goals are sampled much more often than simple goals, leads to the best learning performance in cases where DDPG is combined with for hindsight experience replay (HER). We demonstrate that the approach significantly outperforms standard goal sampling for different robotic object manipulation problems.