https://icelemon.top/tags/%E5%85%B7%E8%BA%AB%E6%99%BA%E8%83%BD/Recent content in 具身智能 on BlogHugozh-CNSat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/embodied-learning-route/Sat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/embodied-learning-route/先建立机器人与学习基础，再完成BC、策略梯度和闭环评测，理解视觉动作模型的接口与研究边界。https://icelemon.top/tech/robotics/embodied-learning-project/Sat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/embodied-learning-project/在可复现的一维闭环中完成合成演示、BC与RL训练、独立评估和梯度检查，并提供PyTorch对照。https://icelemon.top/tech/robotics/imitation-learning/Sat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/imitation-learning/把演示转成观测—动作数据，理解行为克隆的损失、闭环分布偏移和按episode切分。https://icelemon.top/tech/robotics/reinforcement-learning/Sat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/reinforcement-learning/明确状态、动作、奖励和终止，用可运行REINFORCE实验理解回报、基线与闭环评估。https://icelemon.top/tech/robotics/visuomotor-policies/Sat, 24 Oct 2026 00:00:00 +0800https://icelemon.top/tech/robotics/visuomotor-policies/理解动作分布、动作序列和语言条件，检查策略输出的坐标、单位、归一化与执行频率。