5D Library › Books › 5DL-B-043
具身认知、伯恩斯坦自由度问题与运动协同的控制理论
从非受控流形假说、内部正向模型到费茨定律的信息论边界及古典庄子庖丁解牛技进乎道考辨
Embodied Cognition, Bernstein's Degrees-of-Freedom Problem and the Control Theory of Motor Synergies: From the Uncontrolled Manifold Hypothesis and Internal Forward Models to Fitts's Law, Read Beside Classical Chinese Thought on Skill and the Way
DOI10.5281/zenodo.23828057摘要
人的手臂有 7 个关节自由度,把指尖放到空间中一点只需要 3 个;几十块肌肉驱动这些关节,每次动作都可以用无数种方式完成。神经系统怎样在这么多可能性中做出选择?本书用运动控制的数学模型考察这个「伯恩斯坦问题」。依次讨论:伯恩斯坦 1967 年英文版《动作的协调与调节》中的自由度问题与「没有重复的重复」(二手资料引作第 134 页,本书未见原页),格尔方德与拉塔什 1998 年提出、拉塔什 2012 年综述的「丰裕原则」(是对问题的重新表述,不是一个可测的「丰裕度」),以及平面三连杆手臂的示意计算(同一个指尖位置对应肘角 18°—95° 的一整条解曲线;每次换一个解时指尖标准差 6.5 毫米,而同样大小的关节变异若彼此独立则为 121.4 毫米);肖尔茨与舍纳 1999 年的非受控流形(UCM)分析(原研究为 9 人的坐—站任务,方差按每个自由度比较),协同指数 ΔV 与其上下界、对数变换在 ΔV = 0 时不为零、关节坐标与节段坐标给出不同结论(独立噪声的对数比由 0.00 变为 −0.94),以及每次 20 个试次时独立噪声也有 9.4% 的估计超过 +0.5;托多罗夫与乔丹 2002 年的最优反馈控制与「最小干预原则」(只纠正妨碍任务的偏差),在加性噪声的双效应器「求和」任务中,同等纠正努力下任务标准差 1.14 毫米对伺服控制的 2.84 毫米,ln(V_UCM/V_ORT) 为 +4.68,推开一个效应器 10 毫米后另一个反向补偿 4.87 毫米——这是特定模型中的演示,不是「证明」;费茨 1954 年原式 log₂(2A/W) 与麦肯齐的香农式 log₂(A/W + 1) 的区别(16 人、敲击任务最好条件约 10.3—11.5 比特/秒),迈耶等人 1988 年的最优子运动模型中运动时间几乎完全与 √(A/W) 成线性关系(R² = 0.99996,对数式 0.908),以及最小急动度与最小方差是两种理论:后者的速度曲线随肢体刚度从峰值在 0.67T 变到 0.39T,本书开环模型的速度—精度关系更接近幂函数;沃尔珀特、加赫拉玛尼与乔丹 1995 年的感觉运动整合内部模型(没有定位到小脑;小脑内部模型见米亚尔等 1993、沃尔珀特等 1998),延迟约 100 毫秒量级且随通道而异(单突触牵张反射 10—40 毫秒,视觉约 110—150 毫秒),延迟 100 毫秒时卡尔曼估计误差 3.18 毫米对只看延迟读数的 32.37 毫米,未知外力造成约 5 毫米的稳定偏差;布莱克莫尔等 1998 年是 6 人的功能磁共振研究,延迟与旋转的挠痒心理物理见 1999 年,以及舍吉尔等 2003 年每回合平均升级 38% 对应约 0.725 的自感衰减系数(8 回合升高 3.04 牛,原文 3.2 牛);李与承的非负矩阵分解、达韦拉等 2003 年牛蛙踢腿的时变协同、VAF 阈值会低估协同数(真有 4 个时 90% 阈值只选 2 个)、8 块独立控制的肌肉在最小努力解下 4 个成分即可解释 95%、只记录 5 块肌肉时 3 个成分 95.9%,以及库奇与巴列罗—奎瓦斯 2012 年的生物力学约束论证;维赖肯等 1992 年的「冻结—释放」自由度及其反例(纽厄尔与瓦扬古 2001、孔恰克等 2009),一个把冻结看作对内部模型误差的回应的示意模型,沃尔夫的外部焦点研究、蔡等 2021 年元分析(表现 g = 0.264,保持 0.583,迁移 0.584)与麦凯等人校正报告偏倚后的 0.01—0.15(第 10 本已讨论,此处只补充迁移与试验设计的含义),以及一个模型结果:只约束任务无关方向的「内部」指令可以把 ln(V_UCM/V_ORT) 从 +4.69 降到 −1.94 而任务标准差不变(1.11 毫米)。
八章与《庄子·养生主》庖丁解牛(通行本作「臣之所好者道也,进乎技矣」,「技进乎道」是后人的概括)、《庄子·天道》轮扁斲轮「得之于手而应于心,口不能言,有数存焉于其间」以及《列子·汤问》纪昌学射并读,说明它们讲的是养生、言意与技艺传授,不是运动控制理论;最后给出一份四步具身协同核查表(写下任务变量与冗余度、按自由度分解变异并给出区间、检查坐标与零模型、回到任务结果),以及一项预注册随机对照试验方案:在两关节控制一维光标的冗余轨迹追踪任务中,比较「外部任务结果聚焦指令 + 冗余流形反馈」与「内部关节角度固定指令」(唯一主要结局为 24 小时后新轨迹上无反馈的追踪均方根误差,即迁移准确度,以协方差分析校正练习前误差;ln(V_UCM/V_ORT) 为次要结局;最小关注差异 1.5 毫米,标准差 4.0 毫米,相关 0.5,90% 把握度,每组 113 人可分析、招募 133 人,共 266 人,判定规则写成代码)。书中以标注框区分已发表的证据、可运行的 Python 计算(14 段程序,全部核对)、原典与作者的假说;书中所有数据都是虚构或模拟的。本书不是运动医学、物理治疗或康复建议;有伤病或神经系统疾病者应咨询专业人员。
关键词:具身认知;伯恩斯坦问题;自由度;运动冗余;运动丰裕;非受控流形;协同指数;最优反馈控制;最小干预原则;费茨定律;速度—精度权衡;信号依赖噪声;最小急动度;内部正向模型;传出副本;卡尔曼滤波;感觉衰减;肌肉协同;非负矩阵分解;外部注意焦点;庄子;列子
Abstract
The human arm has seven joint degrees of freedom; placing a fingertip at a point in space needs only three, and dozens of muscles drive those joints, so every action can be performed in endlessly many ways. How does the nervous system choose? This book examines this "Bernstein problem" with the mathematical models of motor control. It covers the degrees-of-freedom problem and "repetition without repetition" in Bernstein's 1967 English volume *The Co-ordination and Regulation of Movements* (secondary sources give p. 134; the page was not seen), the "principle of abundance" proposed by Gelfand and Latash in 1998 and reviewed by Latash in 2012 (a reformulation of the problem, not a measurable "degree of abundance"), and an illustrative planar three-link arm (one fingertip position corresponds to a whole curve of solutions with elbow angles from 18° to 95°; choosing a new solution every trial gives a fingertip SD of 6.5 mm, whereas joint variability of the same size, if independent, gives 121.4 mm); the uncontrolled-manifold (UCM) analysis of Scholz and Schöner 1999 (nine people standing up from a chair, variance compared per degree of freedom), the synergy index ΔV and its bounds, a log transform that is not zero at ΔV = 0, joint versus segment coordinates giving different answers (−0.94 instead of 0.00 for independent noise), and how with 20 trials 9.4 % of estimates from independent noise exceed +0.5; optimal feedback control and the "minimal intervention principle" of Todorov and Jordan 2002 (correct only deviations that interfere with the task): in a two-effector "sum" task with additive noise, at equal corrective effort, a task SD of 1.14 mm against 2.84 mm for servo control, ln(V_UCM/V_ORT) of +4.68, and a 10 mm push on one effector compensated by 4.87 mm in the other — a demonstration in one model, not a "proof"; Fitts's original log₂(2A/W) versus MacKenzie's Shannon form log₂(A/W + 1) (16 participants; about 10.3–11.5 bits/s in the best tapping conditions), movement time almost perfectly linear in √(A/W) in the optimised-submovement model of Meyer and colleagues 1988 (R² = 0.99996 against 0.908 for the log form), and minimum jerk and minimum variance as two different theories, the latter's velocity peak moving from 0.67T to 0.39T as limb stiffness rises and the book's open-loop model giving a speed–accuracy relation closer to a power law; the sensorimotor-integration internal model of Wolpert, Ghahramani and Jordan 1995 (not localised to the cerebellum; for cerebellar internal models see Miall and colleagues 1993 and Wolpert, Miall and Kawato 1998), feedback delays of the order of 100 ms that differ by modality (monosynaptic stretch reflex 10–40 ms; visual responses about 110–150 ms), a Kalman estimate with a 100 ms delay erring by 3.18 mm against 32.37 mm for the delayed reading alone, and an unknown external force causing a steady bias of about 5 mm; Blakemore and colleagues 1998 as an fMRI study of six people, the delay-and-rotation tickle psychophysics of 1999, and the 38 % mean escalation per turn of Shergill and colleagues 2003 corresponding to a self-attenuation factor of about 0.725 (3.04 N over eight turns against the reported 3.2 N); the non-negative matrix factorisation of Lee and Seung, the time-varying synergies of d'Avella and colleagues 2003 in kicking frogs, VAF thresholds that undercount synergies (a 90 % threshold picks 2 when there are 4), eight independently controlled muscles whose minimum-effort solutions need only 4 components for 95 %, 3 components reaching 95.9 % when only 5 muscles are recorded, and the biomechanical argument of Kutch and Valero-Cuevas 2012; "freezing and freeing" degrees of freedom (Vereijken and colleagues 1992) and its counter-examples (Newell and Vaillancourt 2001; Konczak and colleagues 2009), an illustrative model in which freezing is a response to internal-model error, Wulf's external-focus research, the meta-analysis of Chua and colleagues 2021 (performance g = 0.264, retention 0.583, transfer 0.584) and the bias-corrected 0.01–0.15 of McKay and colleagues (discussed in Book 10; only transfer and the design implications are added here), and a model result: an "internal" cue that constrains only the task-irrelevant direction can move ln(V_UCM/V_ORT) from +4.69 to −1.94 with the task SD unchanged (1.11 mm).
The eight chapters are read beside Cook Ding carving the ox in the "Nourishing the Lord of Life" chapter of the *Zhuangzi* (the received text says "what I care about is the Way, which goes beyond skill"; "skill advancing into the Way" is a later summary), Wheelwright Bian in the "Way of Heaven" chapter ("I get it in my hand and respond with my mind; my mouth cannot put it into words; there is a knack in it") and Ji Chang learning archery in the "Tang Wen" chapter of the *Liezi*, stating that these concern nourishing life, words and meaning and the passing on of a craft, not motor-control theory. They end with a four-step embodied-synergy checklist (write down the task variable and redundancy; decompose variance per degree of freedom with intervals; check coordinates and a null model; return to the task outcome) and a pre-registered randomised controlled trial in a redundant tracking task where two joints drive a one-dimensional cursor, comparing "external task-outcome focus instructions + redundancy-manifold feedback" with "internal fixed-joint-angle instructions" (single primary outcome the RMS tracking error on a new trajectory without feedback 24 h later — transfer accuracy — by ANCOVA on pre-practice error; ln(V_UCM/V_ORT) secondary; smallest effect of interest 1.5 mm, SD 4.0 mm, correlation 0.5, 90 % power, 113 analysable and 133 recruited per arm, 266 in all; decision rule written as code). Marked boxes separate published evidence, runnable Python calculations (14 programs, all checked), classical sources and the author's hypotheses; all data in the book are fictional or simulated. The book is not sports-medicine, physiotherapy or rehabilitation advice; anyone with an injury or a neurological condition should consult a qualified professional.
Keywords: embodied cognition; Bernstein problem; degrees of freedom; motor redundancy; motor abundance; uncontrolled manifold; synergy index; optimal feedback control; minimal intervention principle; Fitts's law; speed-accuracy trade-off; signal-dependent noise; minimum jerk; internal forward model; efference copy; Kalman filter; sensory attenuation; muscle synergies; non-negative matrix factorization; external focus of attention
目录 · Contents
- 序:同一个动作,无数种做法English 中文Preface: One Action, Endlessly Many Ways
- 第一章 伯恩斯坦问题:冗余、丰裕与「没有重复的重复」English 中文Chapter 1 Bernstein's Problem: Redundancy, Abundance and "Repetition without Repetition"
- 第二章 非受控流形与最小干预原则:只管该管的English 中文Chapter 2 The Uncontrolled Manifold and the Minimal Intervention Principle: Control Only What Matters
- 第三章 费茨定律与速度—精度权衡:信息论类比的边界English 中文Chapter 3 Fitts's Law and the Speed–Accuracy Trade-off: The Limits of an Information-Theoretic Analogy
- 第四章 内部正向模型与传出副本:绕过感觉延迟English 中文Chapter 4 Internal Forward Models and Efference Copy: Getting Around Sensory Delay
- 第五章 肌肉协同:非负矩阵分解与生物力学约束之争English 中文Chapter 5 Muscle Synergies: Non-Negative Matrix Factorisation and the Biomechanical-Constraint Dispute
- 第六章 技能学习:冻结与释放自由度、外部焦点与「指令即代价」English 中文Chapter 6 Skill Learning: Freezing and Freeing Degrees of Freedom, External Focus and "Instructions as Costs"
- 第七章 原典并读:庖丁解牛、轮扁斲轮与纪昌学射English 中文Chapter 7 Reading the Classics: Cook Ding Carving the Ox, Wheelwright Bian and Ji Chang Learning Archery
- 第八章 实践与预注册试验English 中文Chapter 8 Practice and a Pre-Registered Trial
How to cite · 引用
Fan, Y. (2026). Embodied Cognition, Bernstein's Degrees-of-Freedom Problem and the Control Theory of Motor Synergies: From the Uncontrolled Manifold Hypothesis and Internal Forward Models to Fitts's Law, Read Beside Classical Chinese Thought on Skill and the Way [具身认知、伯恩斯坦自由度问题与运动协同的控制理论] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23828057
BibTeX
@book{fan2026embodied,
title = {{Embodied Cognition, Bernstein's Degrees-of-Freedom Problem and the Control Theory of Motor Synergies}: From the Uncontrolled Manifold Hypothesis and Internal Forward Models to Fitts's Law, Read Beside Classical Chinese Thought on Skill and the Way},
author = {Fan, Yafu},
year = {2026},
publisher = {5D Library},
edition = {Version 1.0.0},
doi = {10.5281/zenodo.23828057},
url = {https://doi.org/10.5281/zenodo.23828057},
note = {Chinese title: 具身认知、伯恩斯坦自由度问题与运动协同的控制理论}
}
| DOI | https://doi.org/10.5281/zenodo.23828057 |
|---|---|
| DOI(全部版本 all versions) | https://doi.org/10.5281/zenodo.23828056 |
| ORCID | https://orcid.org/0009-0004-3519-1203 |
| 5D Library ID | 5DL-B-043 |
| 版本 Version | 1.0.0 · 中英双语版 Bilingual edition |
| 出版日期 Date | 2026-10-30 |
| 网址 Online | https://5dlibrary.com/books/embodied-cognition-motor-synergies/ |
| SHA-256 · embodied-cognition-motor-synergies_bilingual_v1.0.0.pdf | 012d35742f3fc4a5dca09b7af42bb345210c8c58a218f5ec97800544c8e101a0 |
| SHA-256 · embodied-cognition-motor-synergies_en_v1.0.0.pdf | a905e741af3834d45f1d77ab3e2660f637fbcb089b6f8b0b8db38336e3cadc77 |
| SHA-256 · embodied-cognition-motor-synergies_zh_v1.0.0.pdf | 4d00c744b8a7d1aed599795c26c579f57b94c3fc99af116e9b6e8a3df65880ce |
| Licence · 许可 | Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0). https://creativecommons.org/licenses/by-nc-nd/4.0/ |
Disclosure. The author uses the works of 5D Library, including this book, as the methodological basis of his community courses, the 5D Ascension Cards (Dimensional Shift Card Decks) and the Yafu AI knowledge base, and therefore has a commercial interest in the subject of this book. The author teaches singing at vocal.guru and singingasia.com and develops movement and perception training courses, and has a direct commercial interest in the subject of this book. The book promotes no course, card deck or service and is not sports-medicine, physiotherapy or rehabilitation advice; the Chapter 8 trial is designed to be run and analysed by researchers independent of the author's business and does not recruit from the author's community, students or customers; all the code and the study protocol are published openly for anyone to check.