RL Training Platform adds Hindsight Experience Replay capability

AI Impact Summary

Hindsight Experience Replay (HER) introduces goal relabeling to improve sample efficiency in sparse-reward environments. Enabling this capability on the RL Training Platform will affect off-policy training pipelines (e.g., SAC, DDPG-like workflows) by expanding the replay buffer with alternative goals and new hyperparameters (goal sampling mode, relabel frequency). Teams should prepare updated training configs, run controlled experiments to compare convergence, and adjust evaluation to account for altered reward signals.