Name: ロボット工学博士だけど「ロボットの未来ついて」質問ある？ | Tech Support | WIRED Japan
Uploaded: 2025-01-24T09:00:45.000Z
Duration: 16 min 20 s
Description: 【看影片學英語】數萬部 YouTube 影片，搭配英漢字典即點即查，輕鬆掌握單字發音與用法，長久累積看電影不必再看字幕。

未來進行式

I'm Henny Admoni. I'm a professor of robotics. Today I'll be answering your questions from Twitter.

Why are people so obsessed with making human-like robots? Y'all need to chill.

I don't disagree. Human-like robots are great. Usually a robot that's not shaped like a human is more robust and more capable of doing the task it needs to do.

One example of a humanoid robot is Sophia the robot. Some people would call this category of robots androids. Sophia is a highly realistic robot, but if you've ever watched Sophia, you know that there's still a gap between what the robot is doing and what we actually expect from people. This is often called the uncanny valley. Very human-like robots can fall into that uncanny valley and make people pretty uncomfortable.

How do you program personality into a robot?

It is hard to define exactly what makes a personality. Robots are just programs.

For example, if somebody makes a funny joke, should the robot laugh or should the robot roll its eyes? This all has to be created in software. When we're programming robots, what we're doing is building a set of instructions. If you run into an obstacle, back up. To get more complex behavior, you can't rely just on pre-specified rules.

That's where something like machine learning comes in. In machine learning, we are building algorithms, which are basically just sets of instructions that can, based on the information that they're getting, adapt themselves to learn the right kinds of output. At christini asks,

Can you pick up a single M&M with a robot arm?

It kind of depends on the robot arm. With this one, it's going to be a little hard.

Some robots are only robot arms, and those are called manipulators, like this one right here.

Ah, so close. This kind of fine motor manipulation is really challenging for robots.

Robots don't have the level of motion and movement that we do in our fingers.

The other thing that we use in our fingers that robots don't have is a sense of touch that allows us to tell when we are actually gripping something so that we can know that it's able to be picked up.

This box has fiducial markers on it. These are essentially QR codes that allow the robot camera to recognize what orientation and position the box is in, supposed to make it easier for the robot to pick up the box autonomously.

Do you think robots will one day take over all of our jobs?

The real benefit of robots is taking over the three Ds, the dull, dirty, and dangerous jobs that we probably don't want human beings to be doing anyway.

People are working on underwater robots that can detect underwater landmines.

Some people have worked on robots that can go into nuclear facilities after an accident and shut off different valves. But I do hope that robots are able to make people better at their jobs and free people up to do things that they're actually good at and that they actually want to do.

What is Elon's goal for the Tesla Optimus robot?

Elon Musk does not consult with me, but my understanding is that he's looking to build a general purpose home robot, one that can do everything from cleaning up your room to unloading your dishwasher.

To design a robot that's capable of doing more than just one thing takes a ton of work and a ton of sophistication.

Tesla is pretty well positioned to do so.

Their autonomous driving division has a huge computer vision capability, and they're building the kinds of batteries that would be needed to power this sophisticated a machine for a long period of time.

What are the major problems in the field of robotics today?

There are a few big challenges that we face today when we're trying to get robots out into the real world.

How do we get robots to see and hear and interact with their environment and actually understand what it is that they're looking at?

How do we get robots to move around in ways that allow them to achieve their tasks?

And finally, another big problem is intelligence.

How do we make our robots smart enough to plan the right kind of action to do in order to achieve what they're trying to?

One major area of interest to me is how we get robots to interact with people, like personal home robots or healthcare robots that are helping nurses and doctors.

I just want to know how Boston Robotics plans to deal with humans who think pushing robots over is hilarious.

There are lots of people who are doing research on how robots should respond to various forms of abuse.

One that I like the most is that when a robot is being kicked by kids in a shopping mall, researchers found that if it just rolls to the nearest adult, the kids will stop.

Have you ever had a genuine social interaction with a robot?

Although I can't say that I've had a deep social interaction with a robot,

I can say that I have felt social reactions to robots.

And that kind of response actually does generate some social interaction for me.

A lot of people are working on social robots to help people in the home or in elder care situations.

And in these cases, it's really important that the robot is able to connect socially with people.

For example, it should tell when somebody's frustrated or not wanting to take a bite of food or somebody's tired and should just go to bed.

Detecting these social cues is still really challenging to build into an algorithm.

And that's a huge part of social robotics.

How hashtag AI is being used in hashtag robotics?

AI is all about how you get systems to act intelligently.

Robotics is about taking devices and putting them out in the real world and having them move around and interact with the environment or with people.

So in order for robots to work, we really need the AI to support them.

But robotics is more than just the AI that supports them.

It's also about the mechanics of building a physical system and the electronics of developing safe circuitry.

Robotics and AI, big overlap, but still two separate fields.

Self-driving cars are already on the road.

When there's a problem, how do we understand the decisions AI makes?

Self-driving cars are essentially a type of robot moving around in the world and making intelligent decisions based on what they're perceiving.

These algorithms that robots are using for decision making are sort of like black boxes.

Inputs come in in the form of perceptions or knowledge, and then outputs come out in the form of actuation.

The robot decides to merge or decides to stop.

But that black box can be very, very complicated.

And because of that, we need something like explainable AI in order to back out the processes that the robot used to get to where it was going.

What types of sensors are used in robots?

Some robots have cameras like this one that can detect not just color images, but also depth using infrared.

Other robots have physical sensors like bump sensors or capacitive sensors that respond to touch, time of flight sensors or other kinds of ways.

The original Roombas also had bump sensors.

That's how they knew they were running into something.

What are swarm robots and how are they learning from insects?

If you watch the Olympics, you've seen these beautiful displays of lights floating in space.

Swarm robots are groups of robots that operate in a collective, have intelligence that's distributed across the entire system so that no one of them has the central information.

These are designed and inspired by the way insect swarms operate, and it's just one example of the ways that robotics has been inspired by biology.

Six-axis robots are called that because they can move in six different directions.

It can move in three directions of physical space, left and right, up and down, and forward and back.

They can also move in three directions of orientation space, pitch, yaw, and roll.

These robots work by deciding how to move every joint in the kinematic chain in order to get the robot hand, which we call an end effector, to the right position.

This process of calculating how the joints will need to move to put the end effector in the right spot is called inverse kinematics.

Do you know who invented the first robot in history?

字幕列表影片播放

ロボット工学博士だけど「ロボットの未来ついて」質問ある？ | Tech Support | WIRED Japan

perceive

sophisticated

physical

impact