Reinforcement learning from human feedback (RLHF) has demonstrated effectiveness in aligning large language models (LLMs) with human preferences. However, token-level RLHF suffers from the credit assignment problem over long sequences, where delayed rewards make it challenging for the model to discern which actions contributed to successful outcomes. This hinders learning efficiency and slows convergence. In this paper, we propose MA-RLHF, a simple yet effective RLHF framework that incorporates macro actions -- sequences of tokens or higher-level language constructs -- into the learning process. By operating at this higher level of abstraction, our approach reduces the temporal distance between actions and rewards, facilitating faster and more accurate credit assignment. This results in more stable policy gradient estimates and enhances learning efficiency within each episode, all without increasing computational complexity during training or inference. We validate our approach through extensive experiments across various model sizes and tasks, including text summarization, dialogue generation, question answering, and program synthesis. Our method achieves substantial performance improvements over standard RLHF, with performance gains of up to 30% in text summarization and code generation, 18% in dialogue, and 8% in question answering tasks. Notably, our approach reaches parity with vanilla RLHF 1.7x to 2x faster in terms of training time and continues to outperform it with further training. We will make our code and data publicly available at https://github.com/ernie-research/MA-RLHF .

Motion planning is a crucial aspect of robot autonomy as it involves identifying a feasible motion path to a destination while taking into consideration various constraints, such as input, safety, and performance constraints, without violating either system or environment boundaries. This becomes particularly challenging when multiple robots run without communication, which compromises their real-time efficiency, safety, and performance. In this paper, we present a learning-based potential field algorithm that incorporates deep reinforcement learning into an artificial potential field (APF). Specifically, we introduce an observation embedding mechanism that pre-processes dynamic information about the environment and develop a soft wall-following rule to improve trajectory smoothness. Our method, while belonging to reactive planning, implicitly encodes environmental properties. Additionally, our approach can scale up to any number of robots and has demonstrated superior performance compared to APF and RL through numerical simulations. Finally, experiments are conducted to highlight the effectiveness of our proposed method.