在杂的机器人优化问题中提高CMA-ES融合速度,效率和可靠性.
Russell M Martin1, Steven H Collins2
1Department of Mechanical Engineering, Stanford University, Stanford, 94305, USA rumartin@stanford.edu.
Evolutionary computation
|January 28, 2026
概括
与标准方法相比,自适应采样CMA-ES (AS-CMA) 通过动态分配评估时间,提高速度和降低成本来优化机器人政策. 这种新的方法提高了噪音环境中的效率,设置的复杂性最小.
科学领域:
- 机器人技术 机器人技术 机器人技术
- 优化算法 优化算法
- 机器学习 机器学习
背景情况:
- 机器人政策优化是耗时的,评估时间会影响速度和准确性的权衡.
- 当前的方法在复杂的优化环境中面临着噪音和效率方面的挑战.
研究的目的:
- 引入自适应抽样CMA-ES (AS-CMA),这是对CMA-ES的增强,以提高优化效率.
- 通过根据预测的分类难度分配采样时间,实现一致的精度.
主要方法:
- 开发了AS-CMA,这是一个新的算法,它补充了CMA-ES与自适应抽样时间分配.
- 将AS-CMA与CMA-ES与静态采样时间和贝叶斯优化在模拟成本环境中进行比较.
- 在现实世界的外骨优化实验中验证了AS-CMA性能.
主要成果:
- 在没有参数调节的情况下,AS-CMA在98%的运行中实现了收.
- 与优化的CMA-ES相比,AS-CMA显示了24-65%更快的收率和29-76%更低的总成本.
- 与贝叶斯优化相比,AS-CMA在复杂的景观中显示出更高的效率和可靠性.
结论:
- AS-CMA提高了优化效率和可靠性,特别是在杂或复杂的环境中.
- 适应性采样策略比静态采样时间提供了实际改进.
- AS-CMA 极小地增加了优化任务的设置复杂性和调整要求.
相关概念视频
Convergent Evolution
32.8K
Evolution shapes the features of organisms over time, ensuring that they are suited for the environments in which they live. Sometimes, selection pressure leads to the rise of similar but unrelated adaptations in organisms with no recent common ancestors, a process known as convergent evolution.
32.8K
Reliability and Validity
13.9K
Reliability and validity are two important considerations that must be made with any type of data collection. Reliability refers to the ability to consistently produce a given result. In the context of psychological research, this would mean that any instruments or tools used to collect data do so in consistent, reproducible ways.
13.9K
Region of Convergence
922
The z-transform is a powerful mathematical tool used in the analysis of discrete-time signals and systems. It is a crucial tool in the analysis of discrete-time systems, but its convergence is limited to specific values of the complex variable z. This range of values, known as the Region of Convergence (ROC), is fundamental in determining the behavior and stability of a system or signal. The ROC defines the region in the complex plane where the z-transform converges, which can take various...
922
Convergence of Fourier Series
400
The Fourier series is a powerful mathematical tool for representing periodic signals as an infinite sum of complex exponentials. In practice, this infinite series is truncated to a finite number of terms, yielding a partial sum. This truncation makes the approximation of the signal feasible but introduces certain challenges, particularly near discontinuities, known as the Gibbs phenomenon.
The Gibbs phenomenon refers to the persistent oscillations and overshoots that occur near discontinuities...
The Gibbs phenomenon refers to the persistent oscillations and overshoots that occur near discontinuities...
400
Distribution Reliability and Automation
507
Distribution reliability in electrical power systems is critical for ensuring an uninterrupted power supply to consumers at minimal cost. According to IEEE Standard Terms, reliability is the probability that a device will function without failure over a specified time period or amount of usage. For electric power distribution, this translates to maintaining continuous power supply and addressing customer concerns over power outages. Several indices, as defined by IEEE Standard 1366-2012, are...
507
Region of Convergence of Laplace Tarnsform
1.1K
The Region of Convergence (ROC) is a fundamental concept in signal processing and system analysis, particularly associated with the Laplace transform. The ROC represents an area in the complex plane where the Laplace transform of a given signal converges, determining the transform's applicability and utility.
Consider a decaying exponential signal that begins at a specific time. When deriving its Laplace transform, the time-domain variable is replaced with a complex variable. This...
Consider a decaying exponential signal that begins at a specific time. When deriving its Laplace transform, the time-domain variable is replaced with a complex variable. This...
1.1K


