请问HN：美学非均匀运动的算法优化
我正在进行自主研究，探索不同数字生态系统中的算法优化。这项工作涉及生成模型，例如用于风格化动画的模型，以及某些居家办公任务自动化的细分领域。
一个反复出现的问题是非均匀运动的建模。标准的运动学模型和路径算法对于基本功能是足够的，但它们无法捕捉到运动中更微妙、几乎是美学的特质，而这些特质对于实现可信度或高保真输出至关重要。最终产生的运动往往显得单调且可预测。优化简单的效率或准确性指标似乎与这个目标背道而驰。
我在寻找不同的方法。是否有特定的架构、目标函数或数学框架在生成或解释故意非均匀且包含细微细节的运动方面证明有效？我对另一种蛮力强化学习的应用兴趣不大，更关注解决这个问题的结构性方案。
查看原文
I'm conducting autonomous research into algorithmic optimization across a few distinct digital ecosystems. The work touches on both generative models, like those used for stylized animation, and certain niches of WFH task automation.<p>A recurring problem is the modeling of non-uniform movement. Standard kinematic models and pathing algorithms are sufficient for basic functionality, but they fail to capture the more nuanced, almost aesthetic qualities of motion that are critical for believability or high-fidelity output. The resulting movement is often sterile and predictable. Optimizing for simple efficiency or accuracy metrics seems to actively work against this goal.<p>I'm looking for different approaches. Are there specific architectures, objective functions, or mathematical frameworks that are proving effective for generating or interpreting movement that is intentionally non-uniform and contains subtle detail? I'm less interested in another application of brute-force RL and more in structural solutions to this problem.