In Proceedings of the Second International Conference on Computational Intelligence, Robotics and Autonomous Systems (CIRAS), Singapore, December 2003.
Challenges and Opportunities of Evolutionary Robotics
D. A. Sofge, M. A. Potter, M. D. Bugajska, A. C. Schultz
Navy Center for Applied Research in Artificial Intelligence
Naval Research Laboratory
Washington, D.C. 20375, USA
{sofge, mpotter, magda, schultz}@https://www.doczj.com/doc/588420590.html,
Abstract
Robotic hardware designs are becoming more complex as the variety and number of on-board sensors increase and as greater computational power is provided in ever-smaller packages on-board robots. These advances in hardware, however, do not automatically translate into better software for controlling complex robots. Evolu-tionary techniques hold the potential to solve many diffi-cult problems in robotics which defy simple conventional approaches, but present many challenges as well. Nu-merous disciplines including artificial life, cognitive sci-ence and neural networks, rule-based systems, behavior-based control, genetic algorithms and other forms of evo-lutionary computation have contributed to shaping the current state of evolutionary robotics. This paper pro-vides an overview of developments in the emerging field of evolutionary robotics, and discusses some of the opportu-nities and challenges which currently face practitioners in the field.
1. Introduction
The field of evolutionary robotics has emerged in recent years as the application of artificial evolution to the de-velopment of robotic systems. While most of the work in evolutionary robotics has focused on the development of control systems for autonomous mobile robots, some re-searchers have used the techniques to evolve robotic hardware configurations and even robot body parts along with the controllers.
Evolutionary robotics (ER) has its origins in several dis-ciplines including artificial life [1], cognitive science and neural networks [2,3,4], behavior-based control [5], ge-netic programming [6] and genetic algorithms [7]. Cur-rent practitioners of ER incorporate techniques from a variety of disciplines to achieve the desired result, often a robotic control system for an autonomous mobile robot (or other autonomous system) which exhibits a set of de-sired behaviors, and for which those behaviors were ac-quired “automatically” (i.e. without custom programming of each individual behavior). The need for ER arises from the fact that as robotic sys-tems and the environments into which they are placed increase in complexity, the difficulty of programming their control systems to respond appropriately increases to the point where it becomes impracticable to foresee every possible state of the robot in its environment. Evolution-ary algorithms are used to generate control algorithms using the Darwinian principle of survival-of-the-fittest. A population of controllers is maintained and evolved by evaluating individual control systems based on a measure of how well they achieve desired characteristics such as executing appropriate behaviors at appropriate times. Only the fitter members of the population survive and pass the characteristics that made them successful on to future generations. The ultimate goal is to produce the best possible controller given some design criteria. As discussed in the following sections, this approach has proven very successful in a wide variety of challenging robotics domains.
The remainder of this paper will provide a sampling of recent development efforts in evolutionary robotics re-search with the goal of showing both the opportunities presented by the use of evolutionary techniques for solv-ing difficult problems in robotics, but also in showing some of the challenges of applying these techniques. 2. Evolved Controllers for Autonomous
Mobile Robots
A key objective in evolutionary robotics is to evolve be-havior-based controllers for autonomous mobile robots [8,9]. Autonomous mobile robots often incorporate both reactive and longer-term planning components in order to accommodate goal-driven behaviors. The reactive por-tion of the controller may be encoded in a variety of forms. Common choices include stimulus-response rules, neural networks, and state-machines. The planning stage may be represented as a series of goal states. Behavior-based controllers are thus driven by a combination of cur-rent state, as determined by current sensor readings and possibly short-term memory, and goal(s). The controller attempts to match its current state readings and goal, which together comprise the stimulus part, and then to produce the appropriate output, or response.
2.1. Evolved Rule-Based Control
Work by Grefenstette and Schultz [10] resulted in the application of the SAMUEL learning system to evolve stimulus-response rules to produce a reactive control sys-tem for autonomous mobile robots. Behaviors achieved using this system include tracking, navigation, and obsta-cle avoidance. SAMUEL maintains a population of can-didate behaviors which are evaluated in a simulated ro-botic environment. The population is scored, mutation and crossover operators applied, and the population is adjusted to remove the lesser scoring rule sets. The proc-ess is run for a number of generations until a stopping criteria is met, at which point the best evolved controller is uploaded to the robot hardware for validation and test-ing. This technique proved successful for evolving reac-tive controllers for autonomous mobile robots. Schultz et al. [12] demonstrated the evolution of rule-based control-lers for learning complex robotic behaviors by evolving the behavior for a shepherd robot to coerce a sheep robot into a corral. This technique and was extended to evolv-ing controllers for simulated autonomous aircraft and autonomous underwater vehicles [7,11].
Challenges with evolving rule-based controllers include determining how to map continuous inputs and outputs to discrete state variables, establishing appropriate interme-diate and goal states, determining how many production rules are required, and performing conflict resolution. 2.2. Evolved Neural Network Based Control
An alternative representation used by many researchers in evolutionary robotics is artificial neural networks, which have a number of characteristics that are desirable from an evolutionary robotics perspective. Neural networks are relatively insensitive to noise in the environment since the output of each node is typically a function of the input from a variety of sources. This characteristic also pro-duces a smooth search space with a well-behaved map-ping between changes to the network and changes in the resultant network behavior. Neural networks also natu-rally handle continuous input and can produce either con-tinuous or discrete output as desired [9].
A key advantage of using evolutionary algorithms for producing robotic neural network based controllers is that the evolutionary techniques may be used equally well with feed-forward or recurrent networks [13]. Recurrent networks offer many advantages for dynamic control sys-tems because they allow recent state information to be combined with current state information in the decision-making process. In effect, the recurrent connections pro-vide a kind of short-term memory capability for the neural network so that decisions may include not only the input state information but also the prior state(s) of the network itself. Recurrent networks, however, are notoriously dif-ficult to train using standard gradient-descent learning techniques [14]. Evolutionary algorithms have proven quite successful in training recurrent neural network-based controllers for autonomous mobile robots [9]. Potter et al. [15] demonstrated the evolution of neural network controllers for multiple robots engaged in shep-herding a sheep robot. The goal of this work was to ex-amine the effects of evolving a single homogeneous con-troller for a group of autonomous mobile robots perform-ing a collective herding task, versus coevolving separate heterogeneous controllers, and to determine if the com-plexity of the task favored homogeneous or heterogene-ous control. The heterogeneous controllers were produced using a cooperative coevolutionary architecture [16] in which the controller for each robot is evolved in a sepa-rate genetically-isolated species. It was found that het-erogeneous controllers are indeed advantageous when the task can be decomposed into subtasks that can be solved by robots specializing in substantially different skill sets. Otherwise homogeneous controllers have an advantage due to their generality.
Quinn et al. [17] described another experiment in which neural network controllers were evolved for a team of real robots. The objective was to study the capability of the robots to learn formation forming behaviors starting from random positions and using a minimal sensor set consist-ing of 4 IR sensors on each robot.
3. Hyper-Redundant Robot Control
One of the greatest challenges in robotics is the design of control systems for robots with high degrees of freedom, particularly if many of the degrees of freedom are cou-pled. A key example of this is a highly segmented ser-pentine or snake-like robotic arm (Fig. 1). This is an ex-ample of a hyper-redundant robot, where there are many possible kinematic solutions to achieve the same end-effector position or trajectory.
Figure 1. Hyper-redundant robotic manipulator Sofge [18] used evolutionary algorithms to generate the inverse kinematics for the hyper-redundant robotic ma-nipulator. The continuous work space of the robot was discretized into a grid of regularly spaced points. A popu-lation of genomes was created such that each genome represented the joint-space configuration of the robot as a real-valued vector. The system then used an evolutionary strategy to coevolve solutions for each grid position
within the workspace such that the distance in joint space
between any two neighboring points was minimized. The fitness function which was minimized included the joint space distance between neighboring configurations, as well as penalty functions for nearing joint angle limits, exceeding the boundaries of the work space, or putting the manipulator into otherwise undesirable positions.
The resulting coevolved kinematics achieved a position error of less than one inch over a simulated 50 ft. manipu-lator while maintaining smooth end-point trajectory con-trol from any point to any other point within the work space.
4. Evolvable Hardware
Evolvable Hardware (EHW) is an emerging field that applies evolution to automate the design of physically reconfigurable structures such as electronic systems, mi-cro-electromechanical systems (MEMS), and robots. Since EHW techniques enable self-reconfigurability and adaptability of the systems with embedded programmable devices, they have potential to significantly increase the functionality, robustness, and reliability of deployable hardware systems. The benefits of self-reconfiguration and adaptation are evident for space exploration, defense, and other applications that need systems to perform with optimal functionality for extended periods of time in un-known, hostile, and/or changing environments. In autonomous robotics this is especially true. Evolutionary algorithms have been applied to the design and post-fabrication adaptation of reconfigurable hard-ware in order to improve the design or performance. The new generation of reconfigurable hardware will also en-able the continuous and embedded evolution of robot con-trollers [19] and vehicle or sensor morphology.
5. Coevolution of Robot Bodies and Brains
In nature, both the morphology and the behavior of an organism are evolved in lockstep. In an effort to explore this more natural form of evolution in ER, a number of researchers have begun work coevolving robot hardware designs with software for controlling the robot. Bugajska and Schultz [20,21] demonstrate the coevolution of both rule-based and neural network-based controllers and sen-sor configurations for autonomous micro-air vehicles (MAVs). The task presented for the simulated MAV is to fly through a forested area to reach a target without hitting any obstacles such as trees. Range sensors with adjust-able beam width and maximum range settings are placed at various locations on the. The SAMUEL system was used for evolving the rule-based controllers and sensor configurations on the same chromosome, while the coop-erative coevolution architecture [16,22] was used for evolving the neural network controllers and sensor con-figurations in two separate species. Successes were dem-onstrated for each method, but more research is needed to better understand the benefits of the use of the coopera-tive coevolution approach over standard evolutionary algorithms, and to determine if the rule-based or neural network-based representation is better for the controller. Karl Sims [23] demonstrated the evolution of fantastical simulated creatures in an artificial reality environment by coevolving the creatures’ minds and bodies. The crea-tures’ bodies consisted of 3D rectangular blocks, while their minds were evolved neural network controllers. The fitness functions were crafted such that the members of the population that could successfully move across the landscape or track a moving target were more likely to survive and reproduce. After coevolution the creatures exhibited a variety of interesting behaviors resembling walking, hopping, slithering and swimming.
Hornby and Pollack [24,25] studied the coevolution of bodies and brains in simulated robots using L-systems as a generative encoding mechanism. The fitness objective was, as with the Sims work, to evolve for locomotion across a simulated landscape. The result of this effort was to demonstrate that the evolved L-system individuals moved faster than those using standard representations, expressed more modularity of design and exhibited a far greater complexity than their simpler brethren.
In the GOLEM project (Genetically Organized Lifelike Electro Mechanics) Lipson and Pollack [26] coevolved simple neural network controllers and robot body parts as a first step towards creating artificial life-forms. The goal of evolution was to learn locomotion behavior. The crea-tures were evolved in simulation, and the body parts were automatically constructed using thermoplastic fabrication equipment. The evolved neural network controller was then uploaded to the artificial neural network hardware. Motors, controller, and body parts were hand-assembled by a human assistant. When the creatures were connected to a power source they demonstrated a crude ability to move across a flat surface such as a table or stage.
6. Embodied Evolution
Embodied Evolution (EE) is based upon the notion that a group of autonomous robots or agents is situated within an environment trying to perform a task. These robots are allowed to interact with one another, mate, and reproduce control programs which then are transferred to other members of the population [27,28]. The key aspects of this paradigm are that the interactions such as mating, cooperation, and so on are based upon contact between the actual robots in the environment, and the entire popu-lation is evaluated in parallel. This inherent parallelism in EE is a major advantage over the more traditional evolu-tionary approach because as the population size increases, the computational requirements for each robot do not. Thus the evolutionary computation scales well, which is
particularly important if it is to be performed on real ro-bots in real time.
Watson et al. [28] describe an early experiment with EE to produce a phototaxis (light-seeking) behavior. Each mobile robot was equipped with two light sensors, two motor control outputs, and infrared diodes to provide omni-directional communications capabilities. The con-trol architecture was a simple neural network model, with inputs from each sensor and outputs to the motors. The evolutionary algorithm used was an adaptation of a steady-state EA model referred to as the probabilistic gene transfer algorithm.
Some of the challenges of getting this system to work were in providing a sufficient number of mobile robots to overcome the stochastic effects of a small population, providing a consistent source of power to the mobile ro-bots, and providing an on-board fitness measure for each robot so that the evolution can be truly distributed. The experiment resulted in a number of solutions comparable to hand-designed behaviors, and a novel looping behavior for reaching the light.
7. Simulation vs. Reality
The most natural and direct application of evolutionary computation to robotics is to perform control-system evaluations on real robot hardware in the actual task envi-ronment. However, if evolution is done in this way a number of issues become problematic. Evolution is a rela-tively long time-scale process that may require many con-trol-system evaluations to achieve satisfactory results, leading to unacceptable runtimes. This problem is exacer-bated by unreliable hardware and poor battery perform-ance. The robot may also enter into dangerous states in which its hardware could be damaged or bystanders in-jured, especially in the early stages of evolution. These issues have led most researchers to evolve control systems in simulation where fitness can be evaluated at faster than real-time speed in a safe off-line environment.
Evolution in simulation is not without its own problems.
A control system that works well in simulation may not perform satisfactorily on a real robot with noisy sensor readings and imprecise motor control while operating a complex environment that is computationally impractical to model with a high degree of fidelity. Grefenstette et al.
[29], in an early attempt to overcome the limitations of evolving control systems in simulation, experimented with adding noise to the sensor models. It was found that given the inevitable mismatch between simulation and reality, it was better to have too much rather than too little noise in the simulation because this encourages more gen-eral solutions to emerge. However, a later study by Jakobi et al. [30] showed that unreasonably high levels of noise can also be harmful because solutions may evolve that rely on the excessive noise. By carefully designing a simulation that differentiates between environmental fea-tures that are critical to the task at hand and those that are irrelevant, the noise can be tailored as appropriate, thus creating control solutions that ignore the irrelevant fea-tures and are robust with respect to the critical ones [31]. Another approach explored by Grefenstette and Ramsey [32] is to combine simulation and reality. In their archi-tecture for anytime learning (also called continuous and embedded learning [33]) control systems are evolved on a simulator that runs continuously on a real robot. Over time, the parameters of the embedded simulation are ad-justed to more closely match reality. As new control rules are evolved that have a high likelihood of improving the robot’s performance, the active robot control system is updated. This process enables the robot to adapt to mis-matches between simulation and reality due to modeling errors, a dynamically changing environment, and changes in the performance characteristics of the robot’s sensors and effectors.
8. Conclusions
The field of evolutionary robotics is growing rapidly. Robots themselves are quickly becoming more complex as the variety and number of on-board sensors increases, and as Moore’s Law results in greater on-board computa-tional power available in ever-smaller packages. Ad-vances in hardware, however, do not automatically trans-late into better software for controlling complex robots. With an increased emphasis on more capable and more autonomous robots, evolutionary techniques hold the po-tential to solve many difficult problems in robotics which defy simple conventional approaches.
Behaviorism has become the predominant paradigm for control of autonomous robots. Artificial evolution pro-vides a means to achieve desired behaviors. Evolutionary techniques also hold the promise of having the robots learn new behaviors automatically in order to adjust to changes in their environments, changes in themselves due to sensor drift or malfunction for example, and to acquire skills for performing new tasks or improving on old tasks. As we develop ever more capable robots which assist humans in performing tasks, it is becoming clear that we must also improve the interfaces and means of interacting with these robots. One approach we are pursuing at the Naval Research Laboratory is known as embodied cogni-tion [34]. In embodied cognition we develop cognitive models of human performance to augment a robot’s rea-soning capabilities. Embodied cognition is based on the premise that use of cognitive models on the robot facili-tates human-robot interaction by making it easier for hu-mans to predict and understand the robot’s behavior and to interact with the robot. This embodied cognition will sit on top of evolved reactive behaviors and will allow future systems to have higher-level cognitive abilities.
For ER to achieve the vision of supplanting the hand-coding and hand-design of complex robotic systems, as pointed out by Mataric [35] it must be shown to reduce the overall effort of its programmers and designers. The evolved controllers produced by current ER systems are often very simple and could be easily bettered by duly considered hand-coded solutions. Representations for robot morphologies and controllers do not often exploit notions of modularity which we expect in more complex creatures. Better representations and better evolutionary algorithms may be necessary to achieve the vision of ER, and better techniques for transferring these solutions into robot hardware, or evolving them in situ, are needed. Acknowledgments
This work was supported by the Office of Naval Research under work order requests N0001402WX20374,
N0001403WR20212 and N0001403WR20057 References
[1] R. A. Brooks, “Artificial life and real robots,” In
Proceedings of the First European Conference on
Artificial Life, MIT Press/Bradford Books, Cam-
bridge, MA, 1992.
[2] D. Cliff, I. Harvey, P. Husbands, “Explorations in
Evolutionary Robotics,” Adaptive Behavior, Vol.
2, No. 1, 73-110, 1993.
[3] J. Barhen, W. B. Dress, and C. C. Jorgensen, “Appli-
cations of concurrent neuromorphic algorithms for
autonomous robots,” In R. Eckmiller and C.v.d.
Malsburg, Eds.,Neural Computers, 321-333,
Springer-Verlag 1987.
[4] R. D. Beer and J. C. Gallagher, “Evolving dynamical
neural networks for adaptive behavior,” Adaptive
Behavior, Vol. 1, No. 1, 91-122, 1992.
[5] R. C. Arkin, “Motor schema-based mobile robot
navigation,” The International Journal of Robot-
ics Research, 92-112, 1989.
[6] J. R. Koza, “Evolving emergent wall following ro-
botic behavior using the genetic programming para-
digm,” In F.J. Varela and P. Bourgine, Eds.,
Toward a practice of autonomous systems.
Proceedings of the First European Conference on
Artificial Life, 110-119, Cambridge, MA, MIT
Press and Bradford Books, 1991.
[7] A. C. Schultz and J. J. Grefenstette, "Using a Genetic
Algorithm to Learn Behaviors for Autonomous Ve-
hicles," In Proceedings of the American Institute
of Aeronautics and Astronautics Guidance,
Navigation and Control Conference (AIAA), 739-
749, 1992.
[8] A. C. Schultz and J. J. Grefenstette, "Evolving Robot
Behaviors," In Fourth International Workshop on the Synthesis and Simulation of Living Systems
(Artificial Life IV), 1994.
[9] S. Nolfi and D. Floreano, Evolutionary Robotics,
Cambridge, MA, The MIT Press, 2000.
[10] J. J. Grefenstette, “Credit assignment in rule dis-
covery systems based on genetic algorithms,” Ma-
chine Learning, 3(2-3), 225-245, 1988.
[11] A. C. Schultz, "Using a Genetic Algorithm to Learn
Strategies for Collision Avoidance and Local Navi-
gation," In Proceedings of the Seventh Interna-
tional Symposium on Unmanned Untethered
Submersible Technology, 213-225, 1991. [12] A. C. Schultz, J. J. Grefenstette, and W. Adams,
"Robo-Shepherd: Learning Complex Robotic Be-
haviors," Robotics and Manufacturing: Recent
Trends in Research Applications, Volume 6, (Eds.
M. Jamshidi, F. Pin, and P. Dauchez), In Proceed-
ings of the International Symposium on Robotics
and Automation (ISRAM '96), ASME Press: New York, 763-768, 1996.
[13] P. J. Angeline, G. M. Saunders, and J. B. Pollack,
"An Evolutionary Algorithm That Constructs Recur-
rent Neural Networks," In IEEE Transactions on
Neural Networks, Vol. 5, 54-65, 1993.
[14] D. Sofge and D. White, (Eds.), Handbook of Intel-
ligent Control: Neural, Fuzzy, and Adaptive Ap-
proaches, Van Nostrand Reinhold, 1992.
[15] M. A. Potter, L. A. Meeden, and A. C. Schultz,
“Heterogeneity in the Coevolved Behaviors of Mo-
bile Robots: The Emergence of Specialists,” In
Proc. of the Seventeenth Int’l. Conference on Ar-
tificial Intelligence, Morgan Kaufmann, 2001. [16] M. A. Potter and K. A. De Jong, “Cooperative Co-
evolution: An Architecture for Evolving Coadapted
Subcomponents,” Evolutionary Computation,
8(1), 1-29, 2000.
[17] M. Quinn, L. Smith, G. Mayley, and P. Husbands,
“Evolving team behaviour for real robots,” In
EPSRC/BBSRC International Workshop on Bio-
logically-Inspired Robotics: The Legacy of W.
Grey Walter (WGW '02), HP Bristol Labs, U.K.,
14-16 August, 2002.
[18] D. Sofge and G. Chiang, “Design, Implementation,
and Cooperative Coevolution of an Autonomous/
Teleoperated Control System for a Serpentine Ro-
botic Manipulator,” In Proceedings of the Ameri-
can Nuclear Society Ninth Topical Meeting on
Robotics and Remote Systems, 2001.
[19] D. Roggen, D. Floreano, and C. Mattiussi, “A
Morphogenetic Evolutionary System: Phylogenesis
of the POEtic Circuit,” In Fifth International Con-
ference on Evolvable Systems, 153-164, 2003. [20] M. D. Bugajska and A. C. Schultz, "Co-Evolution of
Form and Function in the Design of Autonomous
Agents: Micro Air Vehicle Project," In GECCO-
2000 Workshop on Evolution of Sensors in Na-
ture, Hardware, and Simulation, 2000.
[21] M. D. Bugajska and A. C. Schultz, "Coevolution of
Form and Function in the Design of Micro Air Ve-
hicles," In Proc. of 2002 NASA/DoD Conference
on Evolvable Hardware (EH-2002), 2002. [22] M. A. Potter, "The Design and Analysis of a Com-
putational Model of Cooperative Coevolution,"
Ph.D. Thesis, George Mason University, 1997. [23] K. Sims, "Evolving 3D Morphology and Behavior
by Competition," In R. Brooks and P. Maes (eds.),
In Proceedings of the International Conference
Artificial Live IV, The MIT Press, 28-39, 1994. [24] G. S. Hornby, and J. B. Pollack, “Body-Brain Co-
evolution Using L-systems as a Generative Encod-
ing,” In Genetic and Evolutionary Computation
Conference (GECCO 2001), 2001.
[25] G. Hornby and J. Pollack, “Creating High-Level
Components with a Generative Representation for
Body-Brain Evolution,” Artificial Life, 8:3, 2002.
[26] H. Lipson and J. B. Pollack, "Automatic Design and
Manufacture of Robotic Lifeforms," Nature 406,
974-978, 2000.
[27] S. G. Ficici, R. A. Watson, and J. B. Pollack, “Em-
bodied Evolution: A Response to Challenges in
Evolutionary Robotics,” Eighth European Work-
shop on Learning Robots, 1999.
[28] R. A. Watson, S. G. Ficici, and J. B. Pollack, “Em-
bodied evolution: Embodying an evolutionary algo-
rithm in a population of robots.” In Angeline, P.,
Michalewicz, Z., Schoenauer, M., Yao, X., and Zal-
zala, A., Eds., 1999 Congress on Evolutionary
Computation, 335-342, IEEE Press, 1999. [29] J. J. Grefenstette, C. L. Ramsey and A. C. Schultz,
"Learning sequential decision rules using simulation models and competition," Machine Learning, Vol.
5, No. 4, 355-381, Kluwer, 1990.
[30] N. Jakobi, P. Husbands, and I. Harvey, "Noise and
the Reality Gap: The use of Simulation in Evolu-
tionary Robotics," In Advances in Artificial Life:
Proceedings of the Third European Conference
on Artificial Life, Springer-Verlag, 1995.
[31] N. Jakobi, "Evolutionary Robotics and the Radical
Envelope of Noise Hypothesis," In Journal of
Adaptive Behavior, 6(2), 325-368, 1997.
[32] J. J. Grefenstette and C. L. Ramsey, “An Approach
to Anytime Learning,” In Proceedings of the Ninth International Conference on Machine Learning,
189-195, Morgan Kaufmann, 1992.
[33] A. C. Schultz and J. J. Grefenstette, “Continuous and
Embedded Learning in Autonomous Vehicles:
Adapting to Sensor Failures,” In Unmanned
Ground Vehicle Technology II, Proceedings of
SPIE, Vol. 4024, 55-62, 2000.
[34] M. Bugajska, A. Schultz, T. J. Trafton, M. Taylor,
and F. Mintz, "A Hybrid Cognitive-Reactive Multi-
Agent Controller," In Proceedings of the 2002
IEEE/RSJ International Conference on Intelli-
gent Robots and Systems (IROS-2002), 2002. [35] M. Mataric and D. Cliff, “Challenges in Evolving
Controllers for Physical Robots,” In Evolutionary
Robotics, Special Issue of Robotics and Autono-
mous Systems, 1996.
口罩的种类、功能以及使用方法详细讲解 冬季由于天气寒冷、空气污染严重、预防传染病等原因，很多人会佩戴口罩，但是市面上售卖的口罩类型有很多种，佩戴哪种口罩能预防病毒传染？不同的口罩有哪些用途？特殊人群应该如何佩戴口罩？在这个特殊时期，小编就为大家带来了佩戴口罩预防口罩的种类选择和使用方法以及一些其他方面的注意事项，希望对大家有所帮助。 一、戴口罩能预防病毒感染吗？ 常见的口罩有以下几种：普通棉布口罩、一次性口罩（如：医用外科口罩）、N95型口罩。 1、一次性口罩（医用外科口罩） 能在一定程度上预防呼吸道感染，无法防霾。 医用外科口罩有三层，从外到内分别是防水层、过滤层、舒适层。舒适层是一层无纺布，佩戴时白色的无纺布朝内，蓝色的防水层朝外，有金属片的一边朝上，不要戴反，橡皮筋挂上双耳后捏紧金属片和鼻子贴合，抚平两颊，使口罩和面部之间尽量不留缝隙。 购买时要注意，应当选择外包装上明确注明“医用外科口罩”字样的口罩。 2、N95型口罩 能有效预防呼吸道感染，可以防霾。 N95型口罩是是NIOSH（美国国家职业安全卫生研究所）认证的9种防颗粒物口罩中的一种。“N”的意思是不适合油性的颗粒（炒菜产生的油烟就是油性颗粒物，而人说话或咳嗽产生的飞沫不是油性的）；“95”是指，在NIOSH 标准规定的检测条件下，过滤效率达到95%。 N95不是特定的产品名称。只要符合N95标准，并且通过NIOSH审查的产品就可以称为“N95型口罩”。
N95型口罩的最大特点就是可以预防由患者体液或血液飞溅引起的飞沫传染。飞沫的大小为直径1至5微米。美国职业安全与健康管理局（OSHA）针对医疗机构规定，暴露在结核病菌下的医务人员必须佩戴N95标准以上的口罩。 相比较医用外科口罩，N95型口罩密闭更好。在选用N95型口罩时，尽量选择不带呼吸阀的N95型口罩。 3、普通棉布口罩 它的材质可能为棉布、纱布、毛线、帆布及绒等，由于材质本身不够致密，无法起到预防感染目的。 二、如何正确佩戴口罩？ N95型口罩 医用外科口罩
英雄传说空之轨迹FC配置魔法
————————————————————————————————作者：————————————————————————————————日期：
导力魔法表 原表可查看游击士手册。要查阅各种魔法的配置要求，游戏中查看手册即可。此表和手册中的区别在于： 将魔法按照功能分类。 魔法的范围大小更精确。 增加了各攻击魔法的打击数（对于打击数概念的解释，请参考后文【战斗中获得的耀晶片数量与SEPITH UP奖励】部分） 增加了各魔法的驱动时间比。 注：魔法的范围 小圆、中圆、大圆、直线只是比较笼统的说法，大圆有大有小，直线有宽有窄。这里在范围后面补上了一个数字，以表示圆的半径或直线的宽度。例如“直线１”表示直线宽度为１格；“中圆３”表示圆的半径为３格，实际有效范围的直径为７格；“小圆２”表示圆的半径为２格，实际有效范围的直径为５格。 图示如下： 中圆３的有效范围小圆２的有效范围（黑色方块） □□■■■□□□□□□□□□ □■■■■■□□□■■■□□ ■■■■■■■□■■■■■□ ■■■■■■■□■■■■■□ ■■■■■■■□■■■■■□ □■■■■■□□□■■■□□ □□■■■□□□□□□□□□ 范围（非全体）魔法有两种指定范围的方式：目标指定和地点指定。 目标指定：最为常见。只能以某个敌人或队员/NPC为中心。在魔法驱动期间，如果目标移动，魔法的范围圈会跟着移动，如果目标失效（如战斗不能），范围圈会自动换到另一个有效目标上，因此不会发空。单体魔法也可以算是目标指定。 地点指定：圆范围可以指定在战场任何位置，直线可以朝向任何方向。同样范围大小，地点指定比目标制定有可能覆盖更多目标，但是敌人也有可能在驱动时逃出施法范围。所以最好赶在敌人行动前就发出，或者预估提前量，以免浪费。 此表中，范围的数字后面加了“◇”号的，表示此魔法为地点指定。否则为目标指定。
序章父亲、启程 在开篇剧情结束之后（如果不幸黑屏，请参阅置顶或者精品区本人FAQ，还是无法解决请在吧内求助吧友），小约和小艾需要前往洛连特市的游击士协会（PC版界面右上角可查看小地图），在游击士协会与爱娜对话后去二楼与雪拉对话，听过一些基础知识后艾约被带往导力工房，接下来请按照雪拉的要求合成回路并装到主角的导力器上（请至少合成HP1和行动力1，并在艾导力器上装HP1，约导力器上装行动力1）然后给主角中任意一位的导力器开封结晶孔。导力系统介绍完毕后艾约会前往地下水路，开始实地研修。
PS：在协会获得的游击士手册包含游戏大多数基础知识，且记录有游戏主线与支线任务的完成情况，请新手们务必仔细查阅（话说有些童鞋不知道手册能翻页？……）
PSP版的游击士手册、星杯手册为日文版，汉化版见此： https://www.doczj.com/doc/588420590.html,/mxf03/album/%BF%D5%D6%AE%B9%EC%BC%A3-%D3%CE%BB%F 7%CA%BF%CA%D6%B2%E1%2C%D0%C7%B1%AD%CA%D6%B2%E1 ◆主线任务⑴◆实地研修?回收宝物BP：1，500mira 此任务实质为战斗系统教学，请仔细看剧情并按照指示行动 研修结束后，前往里农杂货铺购买《利贝尔通讯》（在买书之前把身上钱全花完的话小约会帮忙付），可获得料理手册并学会枫糖曲奇的制作，并可以开始使用料理系统，料理一般需在酒店饭馆等处购买，也有宝箱开的和任务获得的。 ☆【料理手册】此时可获得的料理：枫糖曲奇、炸薯条(亚班特酒馆、格鲁纳门)、花色苏打（亚班特酒馆）、洋红之眼（亚班特酒馆）、一口气薏粉（亚班特酒馆，大盘料理） ☆收集齐全料理并没有什么特别奖励……没有收集癖的话无所谓，对于新手来说用料理不仅回复效果不错（非战斗状态建议去旅馆或者免费回复点回复，比较省钱），很多料理除了回 复HP外还附带特别效果，做料理也比买药便宜。
功能性绝缘的定义： 3.3.5 functional insulation insulation between conductive parts of different potential which is necessary only for the proper functioning of the appliance。 功能性绝缘的电气间隙的要求： 29.1.4 For functional insulation, the values of table 16 are applicable. However, clearances are not specified if the appliance complies with clause 19 with the functional insulation short-circuited. Lacquered conductors of windings are considered to be bare conductors. However, clearances at crossover points are not measured. 看蓝色字体部分，只要在短路时，能满足clause 19的要求，可不要求functional insulation 的电气间隙。 看红色字体部分，漆包线被视为“裸露导体”。也就是在本标准中，漆包膜不被视为功能性绝缘。这个前提很重要，这意味著绕组间不必进行“短路测试”。 功能性绝缘的爬电距离的要求： 29.2.4 Creepage distances of functional insulation shall be not less than those specified in table 18. However, creepage distances may be reduced if the appliance complies with clause 19 with the functional insulation short-circuited. 看蓝色字体部分，只要在短路时，能满足clause 19的要求，functional insulation 的爬电距离的要求可减小。标准并没有说可以减小到多少，这就意味著，只要functional insulation 能在短路时满足clause 19的要求即可满足爬电距离的要求。 结论：从标准的理解来看，functional insulation 仅为设备正常工作所需的绝缘，如果functional insulation 能在短路时满足Clause 19 的要求，可不要求其电气间隙和爬电距离。 实际上，我也是这么处理的。只要在检查、评估中发现functional insulation 的距离（cl & cr）很小，那么就用短路测试来检验，如果满足不了短路测试，那么我就要求其CL & CR。 我们在日常判断一个器具的功能绝缘其实无处不在，只要存在电势差的两点都可视为功能绝缘，但绝大部分情况下短路都是可以满足Clause 19的要求的，所以我们考核功能绝缘的地方就大大减少了。典型的功能绝缘的例子是L/N输入的地方，发热元件或马达元件的输入端子之间等等 功能绝缘的定义为保障产品功能的绝缘类型,它不像基本、附加、加强绝缘是用来防触电的,也就是说如果这类绝缘距离不够的话,可能会对最终的用户造成触电的危险,而功能绝缘的距离即使不够,用户也不会有触电危险,只不过会造成产品无法正常工作,它也可以用来减少起火的危险.例如电流保险丝之前火线和零线之间或是变压器内置温度保险丝两脚之间的距离
：FC时期导力器很好，5-2链，由于驱动2在FC时期的属性值还算是?不错的，双时限定影响不大，短链正好装攻击，长链则用于配魔法，是最早?能合出炽炎地狱的人，SC导力器降级为5-3链，由于高级回路的出现驱动2?的属性值沦为鸡肋，所以其导力器基本与4-3链无异，还好小约SC,3RD主要?是菜刀，影响也不大 小约可以完全按菜刀回路配，也可以给自己配个强音之力复
配出火山之咆哮和龙卷火焰也是不错的，菜刀的同时偶尔还能砸个魔法 艾斯蒂尔：4-4链无回路限制，看似不怎么样的导力器其实有很多种配法，?比如星之守护，全回复复，暗物质3RD小艾的太极轮是附加伤害仅次于樱花残月的S技，金刚击的伤害也不错?，还是建议不要弄成纯辅助的 小艾的下面这种配法有时改时减、强音复、鬼魅、沉默十字、大地之墙、火?山之咆哮、龙卷火焰、暗物质改等主打魔法，无论是用作辅助还是主攻都是不错的
小艾的纯奶妈回路，主打时改时减全回复，这里银耀是为了出死亡咆哮，银?耀换成封魔/石化之理可出大地之墙，换成红耀可兼顾攻击
满足一些星守控的配法，主打时改时减、强音复、星守、鬼魅、沉默十字、?暗物质改，把机功换成机功2还可出火山之咆哮 奥利维尔：一链，中心孔幻限定，虽然幻限定比较鸡肋，总体来说还是不错?的导力器，但是3RD男性法师由于女性魔装黑丝绸和ATS武器的问题很是悲?剧，于是裸ATS与公主相同，仅次于玲的王子殿下变成了辅助流。。。但是?王子那糟糕的战技论辅助实用度又不如神父，所以。。。真让人同情啊--=
满足一些灾难冲击波控的配法，主打时改时减、强音复、灾难冲击波、虚弱领域、导力停止、龙卷火焰、银色荆棘 传说中的根源屏障（--=无敌鸡肋魔法，普通难度完全无用），主打时改时?减、强音复、星守、导力停止、鬼魅、沉默十字、暗物质改、死亡咆哮，另?外附带范围2
序章父亲、启程 在开篇剧情结束之后，小约和小艾需要前往洛连特市的游击士协会（PC版界面右上角可查看小地图），在游击士协会与爱娜对话后去二楼与雪拉对话，听过一些基础知识后艾约被带往导力工房，接下来请按照雪拉的要求合成回路并装到主角的导力器上（请至少合成HP1和行动力1，并在艾导力器上装HP1，约导力器上装行动力1）然后给主角中任意一位的导力器开封结晶孔。导力系统介绍完毕后艾约会前往地下水路，开始实地研修。
完成情况，请新手们务必仔细查阅（话说有些童鞋不知道手册能翻页？……）
◆主线任务⑴◆实地研修?回收宝物BP：1，500mira 此任务实质为战斗系统教学，请仔细看剧情并按照指示行动。 研修结束后，前往里农杂货铺购买《利贝尔通讯》（在买书之前把身上钱全花完的话小约会帮忙付），可获得料理手册并学会枫糖曲奇的制作，并可以开始使用料理系统，料理一般需在酒店饭馆等处购买，也有宝箱开的和任务获得的。 ☆【料理手册】此时可获得的料理：枫糖曲奇、炸薯条(亚班特酒馆、格鲁纳门)、花色苏打（亚班特酒馆）、洋红之眼（亚班特酒馆）、一口气薏粉（亚班特酒馆，大盘料理） ☆收集齐全料理并没有什么特别奖励……没有收集癖的话无所谓，对于新手来说用料理不仅回复效果不错（非战斗状态建议去旅馆或者免费回复点回复，比较省钱），很多料理除了回复HP外还附带特别效果，做料理也比买药便宜。 ★【红耀石】在研修结束后请尽快与在亚班特酒店左边的民家里的雷特拉对话（此人所在房间内全是书架），可获得《红耀石》第１卷（此卷若错过之后还有次机会购得），若已触发解救孩子的剧情与雷特拉对话就无法获得此书了。 ★此书关系到最终武器的获得，漏掉任何一卷都将无法获得最终武器，若想获得最终武器请收集齐全。 ◆主线任务⑵◆保护孩子们BP：3+1（选择和约修亚一起出击+1），1000mira
强劲入门单反佳能650D功能全面解析（全文） 佳能（中国）有限公司正式发布了全新的EOS 系列3位编号机型EOS 650D。EOS 650D是广受好评的EOS 600D (参数图片文章)后继机型，大幅提升对焦及连拍性能，搭载触控技术的大型可旋转液晶监视器以及众多创意功能，为普及型数码单反相机定义新高度。新型CMOS图像感应器有效像素约1800万，与能高效处理图像并控制相机功能的DIGIC 5数字影像处理器组合，实现了高画质、高感光度低噪点。自动对焦系统搭载了中央八向双十字全9点十字型自动对焦感应器。最高约5张/秒的高速连拍可准确捕捉被摄体，不错过精彩瞬间。 （1/17） 转发到微博 佳能新一代入门级数码单反650D官方产品图
此外，佳能EOS 650D还搭载了实时显示拍摄和短片拍摄时提高自动对焦性能的Hybrid CMOS AF系统，位于图像感应器中央区域的部分像素用于相差检测自动对焦，快速检测合焦位置后，再进行反差检测自动对焦完成精确合焦。沿袭自EOS 600D的3.0″可旋转液晶监视器在EOS系列中首次采用触控面板，带来了直观的操作与便捷的拍摄。短片拍摄模式中搭载了“短片伺服自动对焦”，可以在短片拍摄期间对被摄体进行自动对焦追踪，还能用内置立体声麦克风录音，进行充满临场感、更出色的短片拍摄。另外，支持作品创作的新功能“手持夜景”模式和“HDR逆光控制”模式拓展了表现范围。EOS 650D预计于2012年6月下旬发售。 佳能EOS 650D单反 佳能650D有哪些功能超越了同类入门单反？ ·中央八向双十字全9点十字型自动对焦系统
·最高约5张/秒高速连拍 ·新研发CMOS图像感应器搭载Hybrid CMOS AF实现短片及实时显示拍摄高速对焦 ·支持短片拍摄中持续追踪被摄体的短片伺服自动对焦·搭载触控技术的大型可旋转液晶监视器以及众多创意功能EOS 650D是EOS数码单反相机系列中首款搭载了全十字型自动对焦感应器的3位编号机型。9个自动对焦点都配置了对应F5.6光束的十字型自动对焦感应器，可以不受被摄体形状和拍摄位置影响进行稳定对焦。使用频率高的中央对焦点斜上还配置了搭配大光圈镜头时可发挥效果的对应F2.8光束十字型自动对焦感应器。对应F2.8和F5.6光束的双十字型自动对焦提高了对焦精度。 取景器内的自动对焦点分布 中央八向双十字全9点十字型自动对焦感应器 自动对焦模式设置画面 自动对焦模式有3种，包括适合拍摄静止被摄体的单次自动对焦、动态被摄体追踪能力出色的人工智能伺服自动对焦和可以在单次自动对焦和人工智能伺服自动对焦之间自动切换的人工智能自动对焦。
支线详细图文攻略 作者：cokefish 第一章 ★来たれ食材ハンター[每样各4个以上BP+1] 魔獣の骨 魔獣の牙 魔獣の角 魔獣の鳥肉 魔獣の魚卵 魔獣の鳥卵 单纯的找怪麻烦...只要在海道以及通往学园的街景林道上就可以打全所需的食材。 注: 魔獣の鳥肉..魔獣の鳥卵打通往学园路上的大脚鸟..这两个掉落率似乎不太高..没有幸运回路的话就得多打几次了 此处有机会碰到熊猫..会掉落小艾目前最好的武器...熊猫的攻击力很高..大概2500左右的伤害........ ★ <<紺碧の塔>>の写真撮影[由小艾在正中拍摄BP+1,由朵洛希BP+2] 时间上的限定不是很严苛。可以等到幽灵事件逐渐明了，朵洛希加入队伍时完成。 来到塔顶，有３个点可供拍摄选择，左右两点拍出来的效果最好，不过．．．人家要的是研究照片．．．你的ＢＰ可就看不到附加咯。在拍摄时小艾会停下来问让她来拍是否恰当,选择第2项：由朵洛希来拍。这时候便会看到朵洛希的经典台词“好可爱啊～～～笑一个～～～”了。 当然了，你也可以鼓励小艾去拍，不过即使选择最好的取景点，拍出的也不过是“很不错”的而已，而朵洛希的则是．．．宽屏的．．．１６：９的。ＴＴ同样的相机，专业和业余就差这么多啊。 ★灯台の試運転 你有空的时候（比如做“メーヴェ海道の手配魔獣”任务时），可以顺便到灯塔去看望一下那个固执的老头。 当小艾亲切地和老头打过招呼后正欲出走时，发现他似乎面有难色，便又亲切地接下这个任务（对话选择第一项），原来是老人的腰痛了，无法去操作灯塔，便请小艾代劳了，没办法，不想被人背后说游击士缺少热情，勉为其难了。书架上有说明书，导力灯的具体操作
镜头最全面解析 镜头的全面分析解剖，全面的分析监控镜头，没有比我们更全的描述了，经过几年的搜集整理，今天真功夫安防慢慢分析，摄像机镜头的作用是把被观察目标的光像呈现在摄像机的靶面上，也称光学成像。将各种不同形状、不同介质（塑料、玻璃或晶体）的光学零件（反射镜、透射镜、棱镜）按一定方式组合起来，使得光线经过这些光学零件的透射或反射以后，按照人们的需要改变光线的传输方向而被接收器件接收，即完成了物体的光学成像过程。光学镜头应满足成像清晰、透光率强、像面照度分布均匀、图像畸变小、光圈可调等要求。一般来说每个镜头都由多组不同曲面曲率的透镜按不同间距组合而成。间距和镜片曲率、透光系数等指标的选择决定了该镜头的焦距。 一、镜头分类 摄像机镜头按其功能和操作方法分为常用镜头和特殊镜头两大类。常用镜头又分为定焦镜头（自动和手动光圈）和变焦镜头（自动和手动光圈）。特殊镜头是根据特殊工作环境而专门设计的，一般有广角镜头、针孔镜头等。 二、镜头参数 a、相对孔径 光圈的主要作用是通过控制镜头光量的大小满足成像所需的合适照度。光圈越大，靶面成像照度越大，摄像机输出信号强度越大，信噪比越高。若光圈的实际孔径为ψ，由于光线通过透镜后的折射使镜头的有效孔径D 比实际孔径大，光圈的相对孔径等于有效孔径与镜头焦距之比，即：A=D/f，f 为镜头的焦距。 b、光圈系数 通常将表征镜头光圈大小的参数定义成光圈系数，用F 表示。光圈系数为镜头光圈相对孔径的倒数。F值的规律是后一个值正好是前一个数值的√2 倍，这是由于成像面中心亮度与（1/F）2成正比。F值越小相应灵敏度越大。常用值为1.4、2、2.8、4、5.6、8、11、 16、22等几个等级。 c、视场角 我们常用视场角来表征观察景物的范围。所谓视场角是指在视场角内的景物可全部落入成像尺寸内，而视场角以外的景物将不被摄取。因此，镜头的视场角与摄像机的靶面及镜头的焦距有关。 根据几何原理可以得到视场角的计算公式如下： ωH＝2tg-1（h/2f）ωV＝2tg-1（v/2f） 式中ωH为水平视场角，ωV为垂直视场角，f 为镜头的焦距，h为摄像机靶面的水平宽度，v为摄像机靶面的垂直高度。具体数值可参阅摄像机一节。 当成像尺寸确定后，焦距越短，视场角越大。因此可将镜头分为长角镜头（视场角小于45°）、标准镜头（视场角为45°～50°）、广角镜头（视场角大于50°）、超广角镜头（视场角接近180°）、鱼眼镜头（视场角大于180°）等。 另外，我们还可以得到如下公式以计算其中任一未知项数据。 f/v＝D/V f/h＝D/H 其中：f───镜头焦距；D───镜头至景物距离； h───靶面宽度；H───靶面高度。 三、镜头接口 镜头的安装方式有C型安装和CS型安装两种。C型安装接口指从镜头安装基准面到焦点的距离为17.526mm，而CS型接口的镜头安装基准面到焦点距离为12.5mm。因此将C型镜头安装到CS接口摄像机时需要加装一个5mm厚的接圈。
PC版游戏的第0页有自动存档，如果不幸卡关或者黑屏死机什么的，请读第0页自动存档 地图原尺寸普遍较大，由于度娘会自动缩图，如果觉得看不清楚请点开原图查看 序章父亲、启程 在开篇剧情结束之后（如果不幸黑屏，请参阅置顶或者精品区本人FAQ，还是无法解决请在吧内求助吧友），小约和小艾需要前往洛连特市的游击士协会（PC版界面右上角可查看小地图），在游击士协会与爱娜对话后去二楼与雪拉对话，听过一些基础知识后艾约被带往导力工房，接下来请按照雪拉的要求合成回路并装到主角的导力器上（请至少合成HP1和行动力1，并在艾导力器上装HP1，约导力器上装行动力1）然后给主角中任意一位的导力器开封结晶孔。导力系统介绍完毕后艾约会前往地下水路，开始实地研修。
PS：在协会获得的游击士手册包含游戏大多数基础知识，且记录有游戏主线与支线任务的完成情况，请新手们务必仔细查阅（话说有些童鞋不知道手册能翻页？……）
PSP版的游击士手册、星杯手册为日文版，汉化版见此： https://www.doczj.com/doc/588420590.html,/mxf03/album/%BF%D5%D6%AE%B9%EC%BC%A3-%D3%CE%BB%F 7%CA%BF%CA%D6%B2%E1%2C%D0%C7%B1%AD%CA%D6%B2%E1 ◆主线任务⑴◆实地研修?回收宝物BP：1，500mira 此任务实质为战斗系统教学，请仔细看剧情并按照指示行动 研修结束后，前往里农杂货铺购买《利贝尔通讯》（在买书之前把身上钱全花完的话小约会帮忙付），可获得料理手册并学会枫糖曲奇的制作，并可以开始使用料理系统，料理一般需在酒店饭馆等处购买，也有宝箱开的和任务获得的。 ☆【料理手册】此时可获得的料理：枫糖曲奇、炸薯条(亚班特酒馆、格鲁纳门)、花色苏打（亚班特酒馆）、洋红之眼（亚班特酒馆）、一口气薏粉（亚班特酒馆，大盘料理） ☆收集齐全料理并没有什么特别奖励……没有收集癖的话无所谓，对于新手来说用料理不仅回复效果不错（非战斗状态建议去旅馆或者免费回复点回复，比较省钱），很多料理除了回 复HP外还附带特别效果，做料理也比买药便宜。
红耀石 第1卷 研修结束后，与洛连特民家里的雷特拉对话可得到(我也想不起是那个房子里了,不过好在地方不大)。另柏斯（柏斯超市）有售。 第2卷 一章开始时，与守卫威尔特桥的士兵哈罗德对话可得到(一共就两个兵不是这个就是那个了，好像应该是看门的那个)。另柏斯（柏斯超市）有售。 第3卷 在得知摩尔根将军回来后，与哈肯大门酒吧中的马尔科对话可得到。 第4卷 准备前往瓦雷利亚湖时，去柏斯超市与里布罗对话可得到（就是在超市卖书的那个地方一个新的NPC，好像是在那打工的）。 第5卷 学园祭结束后，去玛诺利亚村前，与卢安市的玛奇尔达对话可得到(一进城坐在桥上的一个女孩)。 第6卷 第三章开始时，去杰尼丝王立学院社团大楼资料室找帕布对话可得到。 第7卷 中央工房事件后，去沃尔费堡垒，与布鲁诺对话可得到。 第8卷 进入终章，返回艾尔?雷登关所，与一楼接待士兵奥塔对话可得到。 第9卷 终章一开始，尽快去格鲁纳门找士兵塞维安对话可得到。 第10卷 剧情到夜晚去大圣堂时，先去空港与在那的拉尔夫对话可得到。 第11卷 武术大会第三天，与王都东区的安敦对话，让他看到中意的女性NPC即可得到。注意场景切换后要重新来过。 集齐全部11卷红耀石后，在进行到集合游击士解放离宫作战的时候，可以去『巴拉尔咖啡厅』，找老板对话。接着就能选择艾丝蒂尔用的「太极棍」或是约修亚用的「黑千鸟?白千鸟」。 序章父亲～起程 (总任务数：7 + 9 = 16，55 BP) （主线）实地研修·回收宝物BP：1 （主线）保护孩子们BP：3 + 1（和约修亚一起出击+1） （主线）剿灭帕赛尔农场的魔兽BP：1 + 2（没有捕捉失败+2/失败一次+1） （主线）克劳斯市长的委托BP：4 （主线）记者们的向导BP：4 （主线）市长官邸的强盗事件BP：6 + 5
PS：PC版游戏的第0页有自动存档，如果不幸卡关或者黑屏死机什么的，请读第0页自动存档 在开篇剧情结束之后，小约和小艾需要前往洛连特市的游击士协会，在游击士协会与爱娜对话后去二楼与雪拉对话，听过一些基础知识后艾约被带往导力工房，接下来请按照雪拉的要求合成回路并装到主角的导力器上（至少合成HP1和行动力1，并在艾导力器上装HP1，约导力器上装行动力1）然后给主角中任意一位的导力器开封结晶孔。导力系统介绍完毕后，艾约会前往地下水路，开始实地研修。 任務主线任务：实地研修 期限- 攻略此任务实质为战斗系统教学，请仔细看剧情并按照指示行动 报酬基本 BP：1 奖励：- 奖金：500mira 【料理手册】此时可获得的料理： 里农杂货铺：枫糖曲奇HP80 亚班特酒馆： 1）炸薯条 HP100 MOV+1 （在以后去杰尼斯王立学院之学园祭的时候也能买到,格鲁纳门卖沙拉三明治的小姑娘也卖这个） 2）花色苏打 HP250 中毒解除（蔡斯市也能买到） 3）洋红之眼 HP100 黑暗解除（蔡斯市也能买到） 4）一口气薏粉（大盘料理） HP180，大盘料理不可以战斗中使用，是平时补充角色用的，比如以后打特务部队长等难度较高的boss前需要以最佳状态出战就需要用到大盘料理了。 收集齐全料理并没有什么特别奖励。没有收集癖的话无所谓，对于新手来说用料理不仅回复效果不错（非战斗状态建议去旅馆或者免费回复点回复，比较省钱），很多料理除了回复HP外还附带特别效果，做料理也比买药便宜。 【红耀石】在研修结束后请尽快与在亚班特酒店左边的民家里的雷特拉对话（最内的屋子），可获得《红耀石》第１卷（此卷若错过之后还有次机会购得），若已触发解救孩子的剧情与雷特拉对话就无法获得此书了。 此书关系到最终武器的获得，漏掉任何一卷都将无法获得最终武器，若想获得最终武器请收集齐全。 【装备回路】武器店能买到： 1）武器：双头棍（艾），苦无（约） 2）衣服：皮革背心，金属背心 3）靴子：皮靴，防滑靴 4）饰品：银耳环，光明背带，黑色手镯 在武器店能买到苦无的话会使接下来的战斗轻松一些，小约双连击可以一回合搞定怪物，如果还有行动加1的回路就更好了。由于游戏初期遇到的魔兽怕火，搞一个攻击１的火系回路也不错。 研修结束后，前往里农杂货铺购买《利贝尔通讯》（在买书之前把身上钱全花完的话小约会帮忙付），可获得料理手册并学会枫糖曲奇的制作，并可以开始使用料理系统，料理一
【图文新手教程】空之轨迹SC流程攻略（附街道/城市/迷宫全地图） 前言 请先玩空之轨迹FC再来游戏，本攻略涉及少量只有玩过FC才会知道的名词，默认所有读者都玩过空之轨迹FC。本攻略主要对应空之轨迹SC的PC版，PSP版的用户也可以参考，两个版本不同的地方我尽量说明。游戏本体的下载、版本问题、OP黑屏问题、存档继承等等问题请到我的空间看一下。 https://www.doczj.com/doc/588420590.html,/gaocheche/blog/item/e81467ee947c68f12f2e21bc.html 本攻略会一直更新，查看最新版攻略请到这里看看（PDF版、EPUB版等其他版本也请到这里下载） https://www.doczj.com/doc/588420590.html,/gaocheche/blog/item/2b938253292e510442a75bbd.html 序章：少女的决意 艾丝蒂尔从晨曦中醒来，新的旅程开始了！人物可控之后直接去天台触发剧情即可，之后是漫长的剧情，再次可控之后从飞艇坪回布莱特家，这段路上会路过洛连特，有爱的同学可以跟各种NPC对话一下，会有很多收获的。到家之后又是剧情，之后序章正式开始。 舞台转移到列曼自治州卢·洛克训练场，人物可控后去二楼房间的床上拿装备，如果有继承SC完美存档的话此时还会得到一块塞姆里亚石，终章可以用来制作最终武器。下楼之后调查亚尼拉丝旁边的椅子坐好，之后会获得新型战术导力器、各属性耀晶片和FC通关奖励(机械手套、圣灵药、土人偶、幸运{回路，提升魔兽的掉宝率30%，刷东西非常有用})，接着跟菲莉斯管理员对话获得料理手册，买点食材做几个“大自然恩惠之水”，这个之后可以用来救命+回复HP，还可以购买《利贝尔通讯第1号》（这个收集了也没用，就是看的，这时不买可以在第八章买到）和料理“香草三明治”。之后去跟维修师罗伯特对话，会获得关于导力器的基础知识，接着合成一些基本回路，建议给小艾一个HP1，亚尼拉丝一个攻击1。准备好了之后离开宿舍，到南边出口跟克鲁茨对话，接着前往SC的第一个迷宫：巴斯塔尔水道。
中式装修风格 中式装修风格一般是指明清以来逐步形成的中国传统风格装修，这 种风格最能体现出中式的家具风与传统文化的审美，造型讲究对称，色彩讲究对比。中式材料以木材为主，图案以龙、凤、龟、狮等为搭配。精雕细琢、瑰丽奇巧。中式装修的墙、地面与普通装修没有什么区别，墙面用白色乳胶漆，也可以用浅色壁纸。地面用木地板、石材、地砖、地毯也都可以。中式装饰材料以木质为主，讲究雕刻彩绘，造型典雅，多采用酸枝木或大叶檀等高等硬木。经过工艺大师的精雕细刻，每件作品都有一段精彩的故事。而每件作品都能令人对过去产生怀念，对未来产生一种美好的向往。 色彩以深色沉稳为主，因中式家居色彩一般比较深，这样整个居室色彩才能协调，再配以红色或黄色的靠垫、坐垫就可烘托出居室的氛围，这样也可以更好地表现古典家具的涵。空间上讲究层次，多用隔窗、屏风来分割，用实木做出结实的框架，以固定支架，中间用棂子雕花，做成古朴的图案。门窗对确定中式风格很重要，因中式门窗一般均是用棂子做成方格或其他中式的传统图案，用实木雕刻成各式题材造型，打磨光滑富有立体感。天花以木条相交成方格形。上附木板，也可以做成简单的环形灯池吊顶，用实木做块，层次清晰，漆成花梨木色。家具设讲究对称，重视文化意蕴，配饰善用字画、古玩、卷轴、盆景、精致的工艺品加以点缀。更显主人的品位与尊贵。木雕画以壁画为主， 更具有文化韵味和独特风格。体现中国传统家居文化 的独特魅力
现代简约装修风格 现代人面临着城市喧嚣和污染，激烈的竞争压力，还有忙碌的工作和紧的生活。因而，更加向往清新自然、随意轻松的居室环境。越来越多的都市人开始摒弃繁缛豪华的装修，力求拥有一种自然简约的居室空间。 风格总述，现代简约装修风格以宁缺毋滥为精髓，合理的简化居室空间。一切从功能出发，讲究造型比例适度，空间结构明确美观，强调外观的明快、简洁。从简单舒适中体现出生活的精致，在材料上首选铁质构件、铝塑板或者合金材料。在设计上，会把机构组织曝露在外，注重室外整个房间的沟通与搭配。在颜色上，多用白色、灰色作为主基调色，并搭配其他颜色的家具，表现个性及章理。风格特点, 墙面、地面、顶棚一集家具设，乃至灯具器皿等均以简洁的造型、纯洁的质地、精细的工艺为其特征。家具突出强调功能性设计，设计线条简约流畅。家居色彩对比强烈，这是现代风格家具的特点。一些线条简单，设计独特极富创意和个性的饰品，都可以称为现代简约风格家装中的一员。 欧式装修风格 提到欧式风格，总让人联想到欧洲宫廷的雅典与精致。它在注重艺术感的同时，又体现了极强的实用性和空间感。欧式风格代表的是 一种生活态度：精致、舒适、豪华、经得起推敲。又带一点点不经意，它更具历史的沧桑感，也许可以流传于几代人的生活，就像那雕刻的时光，高尚、醇厚而又久远。因而，历久弥香。
市场部功能全面解析 企业的销售部门与市场部门是企业营销的两大基本职能部门。市场部门的任务是解决市场对企业产品的需求问题，销售部门的任务是解决市场能不能买到产品的问题，这两个问题同时作用于市场，就是我们今天所做的市场营销工作。市场是企业的龙头，是企业实现产品变成资金、变成利润的主要职能部门；企业生产产品数量、生产产品品种、生产产品规格、产品包装、产品外观等等必须以市场为导向。这些观念已被我们的国内企业接受。有的国内企业已将这些观念在市场实战运用自如，倍尝甜头。 但是，到了20世纪末，一切都在快速发展，于是股东、企业、企业家、职业经理人又觉得这样销售太慢，在寻找诸如“肥猪粉”、“丰乳剂”之类的快速成长添加剂加速销售额、销售量的增长，实现利润的最大化。企业纷纷设立市场部或企划部，进行了第二次营销行动。将广告、策划、宣传、创意、品牌、形象、市场定位等视为神明，认为“广告一响，黄金万两”。有的企业运用市场及广告到了惊天动地的地步，一年之内投入广告一个亿、五个亿已经司空见惯。 纵观百年国际著名消费品公司，都是一步一个台阶、一步一个脚印地发展市场，稳步地增长销售。这些公司有一个共同的营销特征，就是销售与市场二者之间协调的很好，既重视销售，又重视市场，将“拉”与“推”二种力量有机地结合了起来。 下面，本人就根据自己多年市场实战经验，针对市场或企划的基本功能和职责问题加以阐述，并就几个市场部的日常策略设计思路拿出来，供读者参考，也便于心急的企业、心急的人们能清醒地认识市场或企划的真正含义
和作用，以免将市场或企划导入歧途，招人谩骂。 第一部分市场部及各岗位职责 一、市场部的职责 市场部的主要职责有十五大方面。 01、制定年度营销目标计划。 02、建立和完善营销信息收集、处理、交流及保密系统。 03、对消费者购买心理和行为的调查。 04、对竞争品牌产品的性能、价格、促销手段等小的收集、整理和分析。 05、对竞争品牌广告策略、竞争手段的分析。 06、做出销售预测，提出未来市场的分析、发展方向和规划。 07、制定产品企划策略。 08、制定产品价格。 09、新产品上市规划。 10、制定通路计划及个阶段实施目标。 11、促销活动的策划及组织。 12、合理进行广告媒体和代理上的挑选及管理。 13、制定及实施市场广告推广活动和公关活动。 14、实施品牌规划和品牌的形象建设。 15、负责产销的协调工作。 市场部在产品不同阶段侧重点各有不同。 1．在产品导入期，市场部的职责重点有：对消费者购买心理行为的调查；制定产品上市规划；制定通路计划及个阶段实施目标；制定产品价格；制定
英雄传说6全任务BP导览 ================================== 序章父亲～起程 (总任务数：7 + 9 = 16，55 BP) （主线）实地研修·回收宝物BP：1 （主线）保护孩子们BP：3 + 1（和约修亚一起出击+1） （主线）剿灭帕赛尔农场的魔兽BP：1 + 2（没有捕捉失败+2/失败一次+1） （主线）克劳斯市长的委托BP：4 （主线）记者们的向导BP：4 （主线）市长官邸的强盗事件BP：6 + 5（推理正确+4[参考28楼]，唔……只好忍一下+1） （主线）市长官邸的强盗事件②BP：－ （支线）寻找发光的石头BP：2 （支线）牛奶小街的通缉魔兽BP：3 （支线）更换路灯BP：3 + 1 （开锁密码输入成功或交给约修亚来更换+1）（支线）训练士兵BP：3 + 2 （战斗胜利+2） （支线）采蘑菇BP：3 （支线）采集药材BP：3 （支线）寻找小猫BP：2 （支线）艾利兹街道通缉魔兽BP：4 （支线）送亲笔信BP：2 ================================== 第一章消失的定期船 (总任务数：5 + 11 = 16，64 BP) （主线）定期船失踪事件BP：5 + 3 （主线）定期船失踪事件②BP：－ （主线）南街区的强盗事件BP：10 （主线）南街区的强盗事件②BP：－ （主线）南街区的强盗事件③BP：－ （支线）收集食物材料BP：3 （支线）东柏斯街道的通缉魔兽BP：4 （支线）搜寻山道的魔兽BP：4 （支线）迷雾峡谷的通缉魔兽BP：5 （支线）调查熊刺草BP：4 （支线）安塞尔新街的通缉魔兽BP：5 （支线）护卫委托BP：4 + 1 （支线）西柏斯街道的通缉魔兽BP：4 （支线）被盗的戒指BP：3 （隐藏）琥珀之塔的可疑人物BP：4 （隐藏）黑色笔记本BP：5
简介 主人公为成龙.内容是邪恶的大妖术师,无常童子绑架的成龙的恋人明铃,成龙为了救出恋人而斩妖除魔的故事. 游戏画面清新,带有浓厚的中国风味,人物刻画形象生动,人设Q化非常讨人喜爱.表情丰富夸张,动作功夫一招一式也有板有眼,体现出了中国功夫的神彩.比较有创意的就是奖励关了,为游戏添光加彩,难度不是很大,但要想看到礼花也不是很容易的. 优点: 简单明快的横版动作 轻松欢快的BGM 必杀技的描绘 缺点: 不取得ITEM无法发出必杀 Continue画面的意思无法理解 主人公是否以成龙为原形,无从判别出来. 系统介绍 操作 A键跳跃 B键直拳攻击 按住B键闪光时放开飞镖攻击(共5颗)
蹲下时B键扫荡腿 跳跃时B键空中飞脚 必杀技上+B键 根据所加球状物生成不同的必杀技.共有四种: 旋风腿:范围较大的必杀,左右两个方向的敌人都可以攻击到.最大值是7个. 横踢:单个方向的攻击,相对攻击面较小些.但实用性非常强.最大值是9个. 倒勾:只能向上攻击方向攻击,无敌时间几乎没有,但打出来的几率也并不高.最大值是7个. 飞滚:向所朝方向翻滚冲击,拥有最长的无敌时间.最大值7个. p.s. 像空中飞脚或必杀横踢快速左右转向也可以做出两面的攻击. ITEM 游戏中时常遇到青蛙,打中它,就会掉血碗或球状物,不同的球状物有不同的必杀技. Special Stage 游戏里某些地方隐藏着铃铛,(靠近时才能出现,个人认为在左侧较易将其引出),接触后就能进入Special Stage special stage 其中的内容
累积积分,给予加血,人,飞镖(按顺序)的奖励
如果全部踩中(或打中),还会放出礼花. p.s. 在每次Continue之后.仍然会出现各处的铃铛. 特别在Stage4Area4中,每掉落从头开始,都可以进入铃铛中隐藏铃铛的地点: Stage1 Area1 内容:跳云彩 Stage1 Area3 内容:消灭木头人 Stage2 Area1 内容:保护方块
空之轨迹FC——隐藏任务攻略 空之轨迹F C是英雄传说轨迹系列第?部，这部作品可以所?最经典的R P G之?，特别是游戏中的?乐?常有感觉，?且很多玩家也是冲着这款游戏??的?乐来的。 隐藏任务攻略 第?章消失的定期船 1琥珀之塔的可疑?物 委托期限：和将军会?~从王国军那?被释放 报酬：B P42000?拉 经安塞尔新街前往琥珀之塔——打到5层——救下被魔兽攻击的亚鲁?教授 2??笔记本 委托期限：?(发现??笔记本时)~第?章通过古罗尼的?顶关所 报酬：B P52000?拉 在空贼基地B2北侧仓库——调查导?吸尘器发现笔记本——交给哈肯?门地牢的?兵 第?章?花恋诗 1扫荡灯塔的魔兽 委托期限：第?章开始~开始校园?活 报酬：B P41500?拉 到巴伦诺灯塔与弗科特??对话——逐层打上去消灭所有魔兽 2救?任务 在途径梅威海道由刚认识的公主带着去卢安时在沙滩上有?个象“?线天”的地?(就是在?出沙滩的陆地中间有?个可以过去的?缝，在沙滩上时尽量贴?出沙滩的陆地的边?应该能发现)，靠近救的是后来帮他找航海图的吉?，如果救了他在之后的帮他找航海图的任务完成后会有额外的B P 和?拉奖励 3剿灭旧校舍的魔兽 委托期限：在学校遇到乔?和汉斯~学院祭开始 报酬：B P31000?拉 从学校东边的出?到旧校舍——与??处的遇到学??克对话——消灭魔兽(魔兽带毒性攻击，最好装备银?环)
4装饰校园 委托期限：在学校遇到乔?和汉斯~学院祭开始 报酬：B P1500?拉 与学院主楼前的勤务员巴克拉对话——调查以下三个地?——主楼右侧通往社团?楼的?廊——北边男?宿舍前没有挂到旗?处——礼堂?门的右边 5收集资料 委托期限：在学校遇到乔?和汉斯~学院祭开始 报酬：B P1500?拉 与社团?楼2楼的资料室中的?对话——找到放置在以下3个地?的书——社团?楼2楼的男更?室椅?上——教职员室的桌?上——男?宿舍阿吉尔的房间书桌上——交给资料室中的? (以上三个任务是在最后?天排练之后?约他们说要?起吃饭，让?艾和公主去找乔?时完成的，先不要去找乔?，完成以上3个任务以后再去，否则剿灭旧校舍的魔兽会变成?质战提?了难度) 第三章??导?器 1修理运输车 委托期限：完成寻找运输车的?线任务后~提妲开始修理?泵 报酬：B P51500?拉 完成寻找运输车的?线任务后——与中央?房3层设计室的普罗梅笛对话——到5层演算室——通过导?演算机查询运输车的相关情报——到卡鲁迪亚隧道和中央?房的连通处——找??鲁迪拿“驱动?的导?器”——返回运输车事故现场对话 2爱之使者 委托期限：去亚尔摩村修理?泵~提妲开始修理?泵 报酬：B P22000?拉 前往沃尔费堡垒——同守卫?兵布拉姆对话——将情书带给中央?房B1?场的菲。如果在送情书的同时送去礼物会有额外的B P奖励，如果 送“绒?编制帽”奖励B P+4，?王诞?时在望都空港看见的情侣是菲和布拉姆。如果送“果馅饼”或“?作?套 的奖励是B P+2，?王诞?时在望都空港看见的情侣是菲和鲁迪 (以上两个任务都要在进?红叶亭要钥匙之前完成) 如果完成所有?线任务和以上的隐藏任务并在有的任务中会出现的对话选择中选择B P和?拉奖励最?的选项，在离开蔡斯之前就可以成为准游击?1级
PSP空之轨迹FC完整攻略 这游戏画面非常不错，虽说这是老游戏，但是个人认为非常经典，我在以前在电脑上玩了一遍，PSP出了后我又回味一遍感觉还是不错啊！献给喜欢这个游戏的您。 - 1 -
- 2 -
故事开篇，女主角艾丝蒂尔(エステル)在家中等待父亲的归来，此时父亲却怀抱一名受伤的黑发男孩(土产?!)出现在家中，并告知艾丝蒂尔自己将要收养这名男孩。 一转眼5年时间过去，艾丝蒂尔和当年的男孩约修亚(ヨシュア)都已长大成人。这天父亲要求2人前往附近的ロレント城进行准游击士的测试，顺便买一本当前热门的杂志利贝尔通信(リベール通信)最新号回来。两人来到北方城镇内的游击士协会(地图上红色的房子就是)，在2楼见到一位游击士的前辈雪拉莎德(シェラザード)，(因为很熟所以主角们都称她为雪拉姐)。闲聊过后，雪拉告知主角2人参加实地研修，首先来到楼下与柜台姐姐对话，会得到一本手帐，里面详细记录了各种技能的获得方法以及任务的报告书，此后使用V键就可叫出。协会里竖立的扳子上可以看到目前的各种任务，上面写有期限的限制以及高中低难度的划分。 之后在雪拉的带领下来到结晶工房(店名是以老板的名字命名的，懒得打了|||)，从雪拉手中得到各色晶石各一颗，同右上的年轻技师谈话，选择第2项改造·换金，再选第2项进行合成，这里似乎要将所有技能都合成一个，和雪拉对话后将“水”属性的“HP1”技能装备给女主角(选择菜单上的Orbment进行装备)，男主角则装备“时”之属性的“行动力1”。再次和雪拉对话，再对话。然后再同右上技师谈话，这次选择改造·换金后选第1项，对一名人物的技能装备槽进行开封。最后一项实习的内容是前往下水道，在雪拉的带领下(强行带走^ ^)来到入口，进入后就开始打怪吧~。这里的敌人都不强，注意按照提示进行各种战斗方法的实习，最后来到左下角，会发现一个被魔兽守护着的宝箱!打倒敌人后得到小箱x2。将小箱交给雪拉后回到游击士协会，与柜台姐姐对话，选择第2项“报告”，此后每次完成任务后进行报告，就可以获得报酬以及游击士等级的鉴定。再次来到协会2楼，打开刚刚获得的小箱，发现其中所装的是准游击士的纹章，此时2人就真正的成为了准游击士了! 准备回家吧，走出游击士协会的大门后，会看到2个小孩正在嬉闹着说要去秘密基地，对话后往下方城门迈进，经过杂货店的时候别忘记去买父亲要的杂志哦~~。和杂货店的老板对话还可获得料理手帐，按C键即可叫出。出城的时候被游击士协会的姐姐叫住，原来刚才的2个小孩竟然跑去了北方的翡翠之塔!准游击士的第一项任务:保护2个小孩子平安归来。从北面出口进入山道，在第2张地图选择左面岔路一直前进(这里大地图有落差@_@，站在塔下时显示我在矿山~)，来到塔内2层，打倒敌人后救得2个小孩(战斗中要小心保护小孩子，千万不能让他们的HP变0 啊~~)。艾丝蒂尔气愤的教训小孩的时候一只魔兽从后方出现，千钧一发之时父亲出现，一击打倒敌人~。 好了~~这次真的可以回家了。将杂志和在协会得到的信件交给父亲，那封信竟然是帝国方面的联络!晚餐时父亲透露了将要出任务的消息，第2天清早2人为父亲送行……序章正式开始，先到游击士协会，和柜台姐姐对话后接到西方农场治退魔兽的委托。 此时协会看板上的任务有“光る石の搜索”(低)以及“ミルヒ街道の手配魔兽”(低)。 光る石の搜索:与结晶工房左侧门外的人对话，得知他将光る石掉落在杂货店门口，要求主角2人进行搜索，在杂货店门外的排水口发现闪光的东西后，前往礼拜堂后的下水道入口，在左下部分会发现光る石。打退石头引来的魔兽后就去还给石头的主人吧。 ミルヒ街道の手配魔兽:由西侧城门出去后第3张地图的西北方向，可以看到一只比较不同的- 3 -