Hindsight Experience Replay capability added to RL training platform

AI Impact Summary

Introducing Hindsight Experience Replay as a capability expands the platform's reinforcement learning toolbox by enabling relabeling past trajectories with alternative goals, boosting sample efficiency in goal-conditioned tasks. To realize this, environments and replay buffers must support dynamic goal definitions and adjusted reward signals for relabeled experiences, otherwise training may produce misleading gradients. Teams should anticipate changes to data pipelines, evaluation criteria, and memory usage, and may need to tune hyperparameters (e.g., replay buffer size, learning rate) to maintain stability.