Hierarchical Reward Design from Language: Enhancing Alignment of Agent Behavior with Human Specifications
• HRDL extends reward design to encode nuanced human preferences for long-horizon tasks. • L2HR translates natural language specifications into hierarchical reward signals for RL a