探索人类仿生机器人技术：前沿进展与应用

需积分: 9 195 浏览量更新于2024-07-18 收藏 16.2MB PDF 举报

《人形机器人》是由编辑Ben Choi编撰的一本英文书籍，全书共有397页，ISBN号为9789537619442和9537619443。该书涵盖了人形机器人领域的多个重要主题，旨在探索和介绍这一技术前沿的发展。 1. **第1章：人形机器人语言与虚拟现实模拟**（Page 10）——Ben Choi探讨了如何通过虚拟现实技术为机器人提供语言理解和交互的能力，这在构建具有自然对话能力的人形机器人中至关重要。 2. **第2章：基于计算感知理论的情感模仿**（Page 30）——Mohsen Davoudi研究了如何利用计算理论来实现人形机器人的情感表达和模仿，这有助于提升机器人的社交互动性。 3. **第3章：生物基础设计与控制的人形双足机器人**（Page 42）——Giuseppina C. Gini分享了关于仿生设计和控制方法，使得机器人能够更接近人类步态和动态平衡。 4. **第4章：基于归纳学习的语言习得模型的人形机器人连接器获取**（Page 74）——Dai Hasegawa讨论了通过学习模型使机器人逐渐掌握连接和沟通技能，提高其智能水平。 5. **第5章：对“iCub”婴儿人形机器人3 DOF差动式腰部关节性能评估**（Page 92）——W. M. Hinojosa对机器人关节的性能进行了评估，这对于机器人整体运动的精确性和稳定性至关重要。 6. **第6章：基于情绪架构的社会互动机器人**（Page 106）——Jochen Hirth和Karsten Berns探讨了如何设计机器人以理解和回应人类的情绪，从而增强人机之间的互动。 7. **第7章：基于最小二乘法和四次多项式的步行模式生成方法**（Page 126）——Seokmin Hong介绍了如何运用数学优化算法来创建更为自然的人形机器人行走模式。 8. **第8章：强化学习在稳定步行模式生成中的应用**（Page 144）——Lee和Oh展示了如何使用强化学习技术让机器人通过不断试错来优化步行策略。 9. **第9章：反应质量摆动（RMP）模型在人形机器人步态与平衡控制中的应用**（Page 176）——Sung-Hee Lee和Ambarish Goswami探讨了生物力学模型在机器人动态控制中的应用。 10. **第10章：人形机器人凝视控制的神经生理学模型**（Page 196）——Luigi Manfredi深入研究了眼部追踪技术在人形机器人中的应用，模拟人类的视觉行为。 11. **第11章：基于模块化任务结构的动态决策制定**（Page 222）——Giulio Milighetti讨论了如何通过模块化设计帮助机器人在复杂环境中做出明智的决策。 12. **第12章：基于人体解剖和神经分析的人形机器人手臂设计**（Page 240）——Yongseon Moon介绍了如何模仿人类手臂的结构和功能，提升机器人手臂的灵活性和精准度。 13. **第13章：多线性加速度计的6-DOF运动传感器系统**（Page 254）——Ryoji Onodera和Nobuharu Mimura研究了先进的传感器技术在精确测量机器人运动中的作用。 14. **第14章：智能双足人形机器人步态生成方法**（Page 268）——Nizar Rokbani探索了如何通过创新算法生成更为智能的机器人步态。 15. **第15章：具备模仿能力的人形机器人**（Page 282）——Wen-June Wang和LI-PO Chou展示了机器人如何通过模仿学习新的技能和行为。 16. **第16章：可穿戴接口为人形机器人开发新能力**（Page 296）——Hyun Seung Yang等人介绍了可穿戴技术如何增强机器人与用户之间的交互。 17. **第18章：基于心智图像导向语义理论的多代理心灵模型**（Page 342）——Masao Yokota讨论了人工智能在创造更接近人类交流的机器伙伴方面的进展。 18. **第19章：新的人工神经网络控制方法**（Page 374）——Zaier Riadh和Kanda Shinji探讨了如何利用神经网络技术优化人形机器人的运动控制和反射功能。这本书汇集了各领域专家的研究成果，展示了人形机器人技术的最新进展，不仅涵盖了硬件设计、软件算法，还涉及到与人类交互、情感理解和决策制定等高级功能，是了解和研究人形机器人发展的重要参考资料。

Humanoid Robotic Language and Virtual Reality Simulation

4.4 Automatic Constraint Satisfaction

We proposed to use Automatic Constraint Satisfaction to reduce the complexity of

specifying humanoid motions. There are many implicit requirements for locomotion, such

as maintaining balance and structural integrity. Automatic constraint satisfaction system

will provide additional changes to meet the implicit requirements.

A system for providing automatic constraint satisfaction for locomotion is very complex and

much research is being done on areas such as motion planning and constraint satisfaction.

For example, we can simply specify that the robot must move its right hand from current

position (3, 6, 2, 0) to new position (3, 50, 2, 6). The simpler the specification, in most cases,

requires the more complex constraint satisfaction. In our example, the hand must reach the

new position using 6 units of time, so that speeds for various actuators must be adjusted to

meet this requirement. If the hand cannot reach the new position by simply raising it and

reaching out, then the robot must move the whole body toward the new position.

5. Acquiring New Motor Skills

The ability for acquiring new motor skills is essential for mental developments. The trivial

approach is to simply program a humanoid robot for new required motor skills (Ude et al.,

2004), which can easily be done by an experienced programmer using our proposed

language and framework. Thus, in the following we will focus on strategies for acquiring

motor skills through learning from trial and error and learning by macro approach.

Learning motor skills has not yet been a central focus of Machine Learning researchers.

Thus, much research remains to be done on automatic acquiring new motor skills. We

briefly outline strategies for creating such a system, which in part is based on the author’s

work on automata for learning sequential tasks (Choi, 1998; 2003).

5.1 Learning from Trial and Error

One way for acquiring new motor skills is by trial and error. This approach requires first

identifying an objective and then selecting actions or motions to achieve the objective. In

particularly, using our proposed framework, identifying an objective can be described as

identifying a goal state, while selecting actions or motions can be described as selecting a

sequence of transitions from the current state to the goal state.

Using our proposed framework, the underlining model is a non-deterministic finite state

machine (Choi 2002). From one state, there may be several transitions to follow to other

states. Choosing one transitions or the other is why we call this a trial and error approach

and is why it is non-deterministic. To achieve the objective is to find a sequence of

transitions from the current state to the goal state. As soon as a sequence of transitions is

found, it can be stored for future use.

5.2 Learning by Macro Approach

Macro approach can be used in a supervised learning environment (Bentivegna & Atkeson,

2001; Arsenic, 2004; Calinon et al. 2007). After a humanoid robot is repeatedly instructed to

perform certain task, the robot can store the sequence of motions to associate with a name of

the task. After which, the robot can be commanded to perform the sequence by simply

specifying the name of the task. This is a simple record and play back approach.

Humanoid Robots

A more sophisticated approach is provided by the author (Choi, 2002; 2003), in which a

robot can build a non-deterministic finite state machine based on the repeated instructions

to perform certain task. The resulting non-deterministic finite state machine can then be

used for other purposes such as for learning from trial and error as discussed above.

6. Virtual Reality Simulation of Humanoid Robots

We developed a humanoid motion description language, called Cybele, based on the above

proposed framework. Our development process in turn enhances the strength of our

framework. We also developed a virtual reality system to simulate humanoid robots (Zhou

& Choi, 2007). Virtual reality simulation of humanoids has been an important subject due to

its wide range of applications such as motion picture production and robotics (Kanehiro et

al., 2001). To achieve realistic motions, programmers currently must be highly experienced

to define all control factors for the movements of individual body parts on each time

instance. Since it is far too difficulty to control the detail moments of each individual body

parts, currently motion capture systems are widely used to record motions enacted by a

person and then to use the captured motions to animate humanoids (Riley & Atkeson 2000;

Safonova et al., 2002; Nakaoka et al., 2003; Ude et al., 2004).

We attempted a high-level and goal-oriented approach. As shown in Figure 3, there are two

extremes on the approaches to humanoid simulations, the goal-oriented and the motion-

oriented. The goal-oriented approach specifies the objectives using high-level languages,

that are easy to write, and that use Artificial Intelligent methods to generated detailed

motions, but that are much more difficulty in terms of building the system. On the other

hand, the motion-oriented approach can specify detail motion using degree of freedom

(DOF) of each join. However, it is more complicated to write such specifications and

programmers need to provide all the details, but such a system is earlier to be built.

In this chapter we present a new high-level goal-oriented language named Cybele. The

language is designed for humanoid motion description (Choi & Chen, 2002) and is

independent of virtual reality simulation systems. Our new language provides simple

syntax for expressing parallel and sequential processing as well as the combination of these

two. We also introduce syntax for expressing motions. In particular, to solve the motion

mixing problem in parallel and complex blocks, we present an approach to synthesizing

motions with combination of partial states. We define key poses as states or partial states

and create primitives between two key states. We also extract a relatively small set of

parameterized motions from infinite complex motion sequences and make them reusable as

primitive motions. Our simulator interprets the programs, breaks down motions into

primitives with time, scope, and priority. Final motion sequences are then generated using a

synchronization approach.

To achieve constraints satisfaction, we define the scope and priority for each joint in the

humanoid. This system checks constraints on the joints affected by the motion and

determines which motion is in conflict with some others. In addition, to create a multi-

platform system, we implement a prototype system with two parts following the same

paradigm as the Java virtual machine. The first part interprets the program and generates

motion sequences. The second part translates motion primitives to interface with systems

such as virtual reality simulation systems, animation systems, or robotic systems (Kanayama

& Wu, 2000; Hirai et al., 1998; Sony, 2008).

Humanoid Robotic Language and Virtual Reality Simulation

Fig. 3. Two extremes of motion languages

7. Humanoid Robotic Language Specification

Our new humanoid motion description language is called Cybele (Choi & Chen 2002). The

features of Cybele language are object-oriented and scripting motion description language.

Besides the conventional variable declaration, functions, and control flow, we create a new

syntax for describing parallel actions. We also create a new complex motion statement and

expand the use of the conventional motion control concepts using new parallel blocks.

7.1 Motion Statements

To describe humanoid motions, we abstracted a collection of predefined motions for the

whole body and body parts. We decompose the humanoid in a standard way as specified in

humanoid animation specification (Harris et al., 1999). Figure 4 shows the BNF of motion

statements. In general, a motion statement is composed of a main part, parameter part,

followed by a colon ‘:’, a description part, and ends with a semicolon. A main part includes

the object name and the motion name. An object name can be optional. If there is no object

name, the system will take the current default object.

A parameter part takes motion parameters that provide additional detail to specify the

motion. Each expression is separated by comma ‘,’. The description part can be provided

right after the motion function or at the end of a block. It provides a set of attributes to

change the scale of the motion. We currently define two attributes namely starttime and

speed. We can adapt approaches (Pollard, 1999) to retarget and change the scale of motion.

7.2 Specification of Complex Motions

Complex motions can be specified as combinations of sequential and/or parallel sub-

motions. Sequential motions will execute one after another. Parallel motions will execute in

parallel. The start time for parallel motions may not be the same and thus producing

partially overlapping motions.

For ease of specification, we defined our language such that programmers can write

overlapping sequential and/or parallel motions in compounded program blocks. Each block

can be sub-block of other blocks. A compound block can be sequential, parallel, or mixed of

the two.

<motion_statement>

::= <object_serial_option.><motion>‘(’<param_list>‘)’<description_part>‘;’

<object_serial_option>

::= <object> ‘.’| <object> ‘.’ <object_serial_option> | NULL

<param_list>

::= <parameter>| <param_list> ‘,’ <parameter>

剩余396页未读，继续阅读

zztq

粉丝: 7
资源: 53

探索人类仿生机器人技术：前沿进展与应用

人类机器人的研究与应用

"推动机器人基础科技研究的平台-机器人杯研究与实践

"蘇黎世聯邦理工學院機器人技術講義：動力學控制與逆向運動解matlab程序碼

Visual Perception for Humanoid Robots

Humanoid Robots - New Developments

Kicking RoboSapien Humanoid Robots-开源

Humanoid Robots. Human-like Machines

Fast Human Whole Body Motion Imitation Algorithm for Humanoid Robots

人形机器人制造师：Mech Constructor Humanoid Robots 2.0

Anovel compound biped locomotion algorithm for humanoid robots to realize biped walking (2009年)

最新资源