馬新愿ch56人形機器人馬新愿humanoid

1、NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING 1.Background2. Manipulation 3. Activities and Communicatio n 4. Conclusions NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING 1.BACKGROUND The field of humanoid robotics focuses on the creation of robots that are directly inspired by human capabi

2、lities. These robots usually share similar kinematics to humans, as well as similar sensing and behavior. Why humannoids? 1of42 The Human Example The Pleasing Mirror Understanding Intelligence Interfacing with the Human World Interfacing with People Entertainment, Culture, and Surrogates NATURE MACR

3、O CITY FASHION PORTRAIT PHOTO RETOUCHING The human example Roboticists would like to create robots with comparable versatility and skill. When automating a task that people perform, it is natural to consider the physical and intellectual mechanisms that enable a person to perform the task. 2of42 Exa

4、ctly what to borrow from the human example is controversial. The literal-minded approach of creating humanoid robots may not be the best way to achieve some human-like capabilities. For example, dishwashing machines bear little similarity to the manual dishwashing they replace. NATURE MACRO CITY FAS

5、HION PORTRAIT PHOTO RETOUCHING The Pleasing Mirror Humans are humanitys favorite subject. A quick look at popular magazines, videos, and books should be enough to convince any alien observer that humanity is obsessed with itself. 3of42 Robotics serves as a powerful new medium that enables the creati

6、on of artifacts that operate within the real world and exhibit both human form and behavior. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Understanding Intelligence Many researchers in the humanoid robotics community see humanoid robots as a tool with which to better understand humans. Scient

7、ists, developmental psychologists, and linguists have found strong links between the human body and human cognition. 4of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Interfacing with the Human World Many important every day objects fi t in a persons hand and are light enough to be transport

8、ed conveniently by a person. Human tools match human dexterity. For example, humanoid robots and people could potentially collaborate with one another in the same space using the same tools. 5of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Interfacing with People Humanoid robots can potenti

9、ally simplify and enhance humanrobot interaction by taking advantage of the communications channels that already exist between people. 6of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Entertainment, Culture, and Surrogates Humanoid robots are inherently appropriate for some applications. Fo

10、r example, many potential forms of entertainment, such as theater, theme parks, and adult companionship, would rely on a robot that closely resembles a human. 7of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING History and Overview 8of42 Simple humanoid robots Simulating human movements, langu

11、age Integrated visual recognition, speech recognition, bipedal walking Intelligent humanoids Move in a fixed form, cant feel the environment The development of computer technology to realize the sensing, control and control of robots Development of image processing, speech recognition and other tech

12、nologies AI NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Locomotion Bipedal walking is a key research topic in humanoids. Legged locomotion is a challenging area of robotics research, and bipedal humanoid locomotion is especially challenging. 9of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RET

13、OUCHING Bipedal Locomotion Currently the dominant methods for bipedal legged locomotion with humanoids make use of the zero-moment point (ZMP) criterion to ensure that the robot does not fall over. 10 of42 NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING11 of42 If this zero moment falls within yo

14、ur supporting polygon, your body will be able to maintain balance, otherwise you will fall. Bipedal Locomotion ASIMO NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING12 of42 As shown in the figure, in the case of a non- horizontal support surface, the robot will fall even if the ZMP falls within t

15、he support Polygon of the robot. Bipedal Locomotion NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING13 of42 The passive walking theory is proposed to realize the more efficient and natural bipedal walking robot. A large number of human walking biomechanical research results prove that the passive

16、 walking robot is a kind of robot that walks from the inner motion mechanism and becomes a robot. One of the research hotspots in the field. Bipedal Locomotion NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING14 of42 Bipedal Locomotion Why is the 25-second snow- walking episode in the latest versi

17、on of the Atlas video so shocking that Atlas can even do this self- adjustment? NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING15 of42 Figure a) is a schematic diagram of the Capture Region at rest; b) is a schematic diagram of the running time; c) is a schematic diagram of only a few stones in

18、front of the runner when running. The size, shape, and location of the Capture Region are related to your state: you were standing in the N = 0 area when you were standing on one leg, but if someone suddenly pushed you, you instantly had to fall. Speed, it may be that your Capture Region may become

19、N = 1 or even greater than 1. Bipedal Locomotion N-step capture regions NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING16 of42 Why do these humanoid robots have to bend their knees to walk? Their legs are six degrees of freedom, and when the knees are straight, the mechanical legs are in the sin

20、gular state that we are talking about. Bipedal Locomotion NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING17 of42 Aside from the complexity of the body structure itself, the legs and arms are inherently different ways of working. For this requirement, many foot robots use a special actuator calle

21、d the Series Elastic Actuator (SEA, Series Elastic Actuator), which is a design with a spring in series with the motor and gearbox. Bipedal Locomotion Schematic diagram of HRP-3 humanoid robot NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING18of42 It is necessary to be able to measure such a larg

22、e force, and to make it smaller, the spring is only very hard (Stiffness is large enough), and the hard spring is inevitably very large. So, the legs that actually use SEA are like this: Bipedal Locomotion NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING19 of42 The actuators used by Boston Dynami

23、cs are not motors at all, but hydraulic systems. The hydraulic system has two important advantages over the motor system: high energy density and high energy utilization efficiency. The high energy density means that the hydraulic system can output more force than the motor system under the same sys

24、tem volume; the high energy utilization efficiency means that the energy loss of the hydraulic system is smaller than that of the motor system. Bipedal Locomotion NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Falling Down 20 of42 A human-scale robot should expect to fall from time to time in r

25、ealistic conditions. A humanoid robot may fall down due to a large disturbance even if the motion is planned carefully and a sophisticated feedback controller is applied to the robot. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Localization and Obstacle Detection 21 of42 Hondas Asimo uses a

26、camera mounted on its lower torso that looks down at the floor to find artificial markers for position correction Disparity images generated by stereovision have been utilized for this purpose. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING 2. Manipulation 22 of42 Manipulation research within h

27、umanoid robotics typically focuses on the use of anthropomorphic arms, hands, and sensors to perform tasks that are commonly performed by people. The kinematics of humanoid robot arms emulate the human arm, which can be approximated by seven degrees of freedom (DOFs), with three at the shoulder. The

28、 humanoid robot Justin has two seven-DOF torque-controlled arms and two 12-DOF hands . NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Sensing for Manipulation 23 of42 Asimo uses a binocular stereo vision similar to humans to run a visual algorithm of his own secrets. For robots moving outdoors

29、or in more complex environments, obtaining accurate 3D terrain information in advance is important for achieving stable biped walking control. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Sensing for Manipulation 24 of42 LIDAR analyzes the three-dimensional terrain of the scene by transmittin

30、g and receiving the reflected laser light in all directions of the environment. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Sensing for Manipulation 25 of42 The LIDAR used on the head of a humanoid robot can usually detect tens of meters away while achieving centimeter accuracy in the distan

31、ce of a millimeter. The world it sees is probably like this: NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING26 of42 A. Model-Based Vision Sensing for Manipulation B. Feature-Based Vision A common approach to visual perception for humanoid robotsisreal-time three dimensional(3-D)object recognitio

32、n and six-DOF pose estimation using models of the objects to be manipulated. In order to generalize a task across different objects, only the task-relevant features need to be detected and mapped. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING Learning and Development 27 of42 Goal Driven Non-Go

33、al Driven Sensing for Manipulation Encoder - Decoder Frame Semantic coding word order : NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING28 of42 Encoder the input sentence X and converts the input sentence into an intermediate semantic representation C by nonlinear transformation: The word yi to b

34、e generated at time i is generated according to the intermediate semantic representation C of the sentence X and the historical information that has been generated before: Learning and Development Sensing for Manipulation NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING29 of42 The Encoder replace

35、s the RNN model with a multi-layer forward neural network. The output of the neural network represents the intermediate semantic representation of the context information Context and the current input Message, and the Decoder generates a conversation response based on this intermediate representatio

36、n. Learning and Development Sensing for Manipulation NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING30 of42 The figure shows an overview of several approaches to whole-body motion generation for humanoid robots. These methods decouple the problem by computing a coarse movement and transforming i

37、t into a d y n a m i c a l l y b a l a n c e d movement. Then, during the whole-body activity, the humanoid uses a sensory feedback control system to s t a b i l i z e i t s m o t i o n . 3. Activities and Communication NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING31 of42 There are currently f

38、our prevalent ways to generate coarse whole-body motion: 3. Activities and Communication Consequently, human motion is often used when generating the motion of a humanoid robot. Motion capture systems, such as those used by movie studios for special effects, are commonly used for recording human mot

39、ion. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING32 of42 Fast path-planning techniques such as rapidly exploring random trees (RRT) can compute collision-free full body motions for humanoid robots automatically .Given geometric models of the humanoid and the environment, an initial posture, a

40、nd a goal posture, the planning system automatically searches for a collision free whole-body motion that moves the humanoid from the initial posture to the goal posture 3. Activities and Communication NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING33 of42 The dynamics filter proposed by Yamane

41、and Nakamura can convert a physically infeasible motion into a feasible one for a given humanoid. This consists of a controller and optimizer as shown. 3. Activities and Communication Generating Dynamically Stable Motions Introduce two methods to transform rough motion into dynamic and stable motion

42、. Dynamics Filter NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING34 of42 The figure shows motion-captured data of a human walking and the resulting, physically feasible motion produced by the dynamics fi lter. The motions look similar, but the contact conditions for the feet have been modifi ed

43、to be physically feasible. 3. Activities and Communication NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING35 of42 Autobalancer 3. Activities and Communication The autobalancer calculates all joint angles at every sample in time by solving a quadratic programming optimization problem in order to

44、convert a given motion to a balanced one. The autobalancer calculates a whole-body motion based on the following concepts. 1.Fix the center of gravity (CoG) on the vertical axis whichpassesthrougha pointinthesupportpolygon of the humanoid; 2.Keepinertiamomentsaroundt heCoGatacceptable values in orde

45、r to satisfy the balancing conditions. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING36 of42 Generating Whole-Body Motions from Operational Point Motions 3. Activities and Communication The example GUI and teleoperation interface described previously, enable a person to control the motions of p

46、articular points on a robot. Likewise, the autonomous control of real robots with nontrivial dynamics requires real-time methods. NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING37 of42 3. Activities and Communication T h i s c o o rd i n a t i o n o f controllers uses three distinct control cate

47、gories: constraint handling, operational tasks, and postures. Generating Whole-Body Motions from Operational Point Motions NATURE MACRO CITY FASHION PORTRAIT PHOTO RETOUCHING38 of42 3. Activities and Communication A ladder-climbing behavior tested in simulation demonstrates this framework for control.