Week 3
“Introduction … Video Lecture … Weekly Assignment (W3)”
Summaries
- Week 3 > Video Lecture > Video 1
- Week 3 > Video Lecture > Video 2
- Week 3 > Video Lecture > Video 3
- Week 3 > Video Lecture > Video 4
Week 3 > Video Lecture > Video 1
- First of all, I would like to explain about what I’m going to talk about by utilizing- using very simple explanations.
- The left-hand side, we saw the- what they call computer-controlled vehicle, which is very high performance vehicle.
- We don’t have to apply any actuation via motors and so on.
- The left-hand side is very similar to the conventional type of robots.
- I mean, normally, the conventional type of robot is controlled by electric motors.
- On the other hand, all these- in this- the robot, the conventional robot, all motors are precisely controlled.
- If we utilize the robot dynamics property, like a glider, we can decrease the amount of computation.
- If the robot is designed appropriately, so that it can walk, it can utilize these dynamics.
- Let’s look at some examples of the robot, so that we want to see what is the meaning of utilizing dynamics.
- For the conventional robots, the jumping is a very difficult task because it would have to control every motor very precisely.
- The dynamics of the robot is properly designed so that it can jump.
- As a result, the robot needs less computation for realizing jumping.
- It is a really complicated musculoskeletal structure, which has several muscles- we can see nine muscles in this figure- and also bones.
- So let’s look up the jumping of the robot in real.
- So it’s like a glider- less computation and very natural motion.
- Additionally, what is really important for intelligent behavior is compliance.
- So to understand the intelligent behavior, one approach is, of course, building human-like anthropomorphic body and testing this hypothesis that we have on our intelligence.
- First, I would like to talk about the compliance of muscular-skeleton system, then dynamics of the muscular-skeleton system, finally going to the function of the elastic skin, which is also the dynamic property of our skin.
Week 3 > Video Lecture > Video 2
- If you put the plastic sleeve on the rubber tube the growth of the volume is converged to the longitudinous force to pull the things, to move the robot.
- By utilizing such pneumatic muscles- because the pneumatic muscles is made by rubber tube, it is, basically, very compliant.
- What we need to control the robot is, we need two actuators, two pneumatic muscles, to control one joint, which we call push-pull.
- If you change the tension of the muscle, you can change what kind of amount you introduce by the muscular-skeletal system to the interaction between the environment and the body.
- So it is very important and very interesting nature of muscular-skeletal system.
- In these figures, I show several types of robots we developed in our laboratory.
- Each robot is trying to mimic human-like muscular-skeletal structure.
- This is one of the robots, which is, basically, upper arm and shoulder.
- Anthropomorphic means human-like, human-like robot arm, which has a soft hand and seven degrees of freedom robot arm.
- It’s not really looks like a seven degree of freedom robot arm.
- I mean, the conventional robot which has seven degrees of freedom has seven motors.
- In the hand, we have some touch sensors in it, and we use the conventional strain gauge in the silicone skin, so it’s very compliant.
- The strain gauge is installed to some artificial muscles, which sense the expansion of the muscle itself.
- So we also have touch sensors in the hand, and also have some proprioceptive sensors in the muscles, so that we can sense the state of our body.
- What is very interesting is that the structure of the compliance is really similar to the human.
- So let’s look at the example of the dynamic touch by the robot.
- Now, the robot is grasping a plastic bottle(PET bottle).
- Now, the robot is trying to figure out what kind of contain in the bottle.
- This is an experimental result, how the robot feel the contents.
- In these figures, we can see that, if you shake the object in a vertical way, then it is very sensitive to the weight of the contents.
- In this robot, we try to realize, as much as possible, not only the arm structure but also the shoulder structure, so that we can try to imitate the whole upper body, upper arm motion.
- In this robot, what is very interesting feature is the shoulder blade, what we call.
- We have a shoulder blade, here, and that shoulder blade is very interesting structure of the human.
- So the range of motion is really wide, and it’s very interesting how it will move, when you lift your arm up.
- The right-hand side of bottom figure, it says, the very typical motion of the shoulder blade.
- We put muscles on the robot, and then we can realize almost the same range of motion of the shoulder as a human.
- So it is very interesting to know that, even if you don’t have muscle to lift your arm up, the shoulder blade will lift your arm up.
- This is one example of the movement of the robot, which is throwing.
- Throwing is also very interesting behavior of a human.
- Because, if the body is very stiff, you cannot throw a ball so far.
- If the body is very soft, you can utilize flexibility to throw a ball farther.
- Because the structure of the robot is very similar to a human, realizing human-like motion is a little bit easy.
- You can see that the movement of the robot is very similar to a human.
- Opening door is actually very difficult task for conventional robot, because, if you want to open the door by a robot, you have to precisely control the robot.
- Because the conventional robots are really, really hard, it will likely break the door, if you don’t apply any control.
- Since this robot is built by compliant muscles, basically, the robot is very flexible.
- So it don’t have to control the motion of the hand very precisely instead, just grab it, and just turn the knob, and just open it.
- The idea is, if the robot has a similar structure as a human, the robot is supposed to have a similar characteristic, not only the aesthetic, but also the dynamic characteristics.
- It looks like exploring the door, right? But, actually, we apply only random input to the robot, which means that, even if we apply the random robot, the structure of the muscular-skeletal system will generate human-like motion.
- So this is very, very interesting example to demonstrate, structure and compliance is very important to realize human-like behavior, like opening door.
Week 3 > Video Lecture > Video 3
- Well, in the conventional robots, they basically utilize gears and motors.
- So this robot contacting to the ground and it will change the shape because of the compliance of robot.
- These are the diagrams when we control the muscle and when we release the force from the muscle.
- So basically that kind of jumping motion can be achieved by tuning the timing between the muscles.
- Now he’s tuning the timing between the muscles so that it can jump upward.
- Otherwise, if the timing is not appropriate, then the robot will moving forward and moving backward and cannot really maintain jumping.
- Human has basically nine muscles to control one leg.
- What is very interesting is that now look at the muscle five, six, and nine.
- Look at muscle five is attached to the body to the under part, thigh, of the leg.
- So the muscle five controls not only the waist joint, but knee joint.
- The muscle nine is connecting from thigh to the ankle.
- The muscle nine is controlling knee and ankle simultaneously.
- Means that if we change the excitation of these muscles, we can change the coordination between joints.
- That we see that these muscles, what we call biarticular muscles, which control not only one but two joints simultaneously.
- Because of that, the coordination between the joint is controlled by that kind of muscle, we actually don’t need too much computation, as I said.
- Also controlling the muscles we don’t need very precise motion.
- For sensing, sensing is more interesting because if we want to control the conventional type robot then we need a lot of sensors to control the motion of the robot.
- For this robot, because the coordination is kept by the structure of the muscles, we only need one tiny pressure sensor which senses the touch against the ground.
- In this slide, I would like to explain how the coordination is kept by biarticular muscles.
- Imagine that we extend the knee by muscle number three.
- The muscle number three is actually monoarticular muscle to extend the knee.
- If you don’t have any biarticular muscles, then we- just apply the excitation of muscle three will end up the motion just of the motion of the knee.
- Since we have biarticular muscles nine and six, then the- for example, muscle six is pulled by the knee so that it can extend the upper torso as well.
- Muscle nine also will pull and the ankle is extended like this.
- So by just applying the excitation to the muscle three, we’ll end up to the whole body motion to stretch not only knee but upper torso and ankle as well, like this.
- So if we- of course, it is very easy to understand that if you change the excitation of the muscles six and nine, biarticular muscles, so that we can change the coordination.
- This is what we get from the behavior of the robot by changing the tension of the muscle nine, which is called the gastrocnemius.
- So if we loosen this muscle, the robot will go forward.
Week 3 > Video Lecture > Video 4
- It has bones, and also the bones are driven by some tendons, like a human, and also the bone is covered by two types of- two layers, I should say- two layers of elastic skin, and in these two layers, we put several sensors so that we can imitate the sensing ability of a human.
- Imagine that if you are blind- if you get some and your eye is covered, and you’ve got some object, and then try to figure out what you’re graphing, maybe you will do exploring on the object by moving the hand, right? So that kind of behavior we call repetitive grasping so that we can estimate what you are grasping by exploring the object.
- What is a very interesting feature of this repetitive grasping is even if you start from different posture, like left hand side, by applying the repetitive grasping, the object will gradually move to a kind of stable position between the hand and the object.
- Left and right videos- we start quite different posture to start, but through the repetitive grasping, it will gradually move to a stable position, and as a result the grasping looks very similar.
- Before we apply the repetitive grasping, say, in the graph, the triangle, and the square, and the circle, and the cross indicate a similar object, but starting from different posture.
- Throughout these experiments and the examples, you can see that the goal of the research- first of all, in the beginning of the lecture, I said that we really want to understand the human-like intelligence by realizing the human-like body, and also we really want to estimate the methodologies to building up a humanoid with such kind of human-like abilities.
- As we see, pneumatic muscles and also the elastic skin are key features to realize human-like compliant body, and by gradually increasing the complexity, we can imitate human behaviors, and as a result, we hope we will realize human-like- really human-like humanoid.
