Scholarlink

An improved Preference-Based Reinforcement Learning (PbRL) framework for autonomous systems is presented in this research. By incorporating an initial trajectory collection process, the framework significantly reduces data collection time, improving training efficiency. Through human preference feedback, agents iteratively refine their policies without explicit reward engineering. Experiments in simulated driving scenarios demonstrate that the agent produces human-like trajectories, closely aligning with human preferences while prioritizing safety and adaptability. The results highlight the effectiveness of the modified framework in advancing decision-making for autonomous systems, bridging the gap between machine-optimized behaviors and human-desired outcomes.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Enhanced Preference-Based Reinforcement Learning Framework for Autonomous System

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Dake Ding,
  • Raihan Kabir,
  • Nnadi Leonard Chukwualuka

摘要

An improved Preference-Based Reinforcement Learning (PbRL) framework for autonomous systems is presented in this research. By incorporating an initial trajectory collection process, the framework significantly reduces data collection time, improving training efficiency. Through human preference feedback, agents iteratively refine their policies without explicit reward engineering. Experiments in simulated driving scenarios demonstrate that the agent produces human-like trajectories, closely aligning with human preferences while prioritizing safety and adaptability. The results highlight the effectiveness of the modified framework in advancing decision-making for autonomous systems, bridging the gap between machine-optimized behaviors and human-desired outcomes.