Robotics and humanoid robotics & civilian drones discussion

broadsword

broadsword

Brigadier
I can't wait for the next stage of robotics after the lower limb control - hand and finger dexterity. That is a bigger challenge than endurance running. Then we will know how close a robot is to taking over human chores like vacuuming the house using a traditional vacuum cleaner and cooking fried rice.
 
G

GulfLander

Major
Registered Member
"A Shanghai-based robotics firm has unveiled a groundbreaking open-source dataset that it hopes will allow companies and research institutions around the world to better train robots in replicating the nuanced dexterity of human hands.

The dataset was built using a revolutionary VR-driven method: humans wearing headsets guided a humanoid robot's hands via teleoperation, capturing intricate real-world movements."
 
S

SlothmanAllen

Junior Member
Registered Member
I feel like all of these humanoid robots are still very early and each on of these companies kind of has a small piece of the overall puzzle that they are specializing in. I think it will be a while before one of these companies is able to bring it all together in one package, or maybe we see various humanoid robots of different configurations in the end. Either way, I think it is very early days for this technology.

Here is cool video so some Figure .02 robots collaborating on a task.

 
I

iewgnem

Senior Member
Registered Member
PeoplesPoster said:
Eh I wouldn't say that, this latest video's movements are quite a bit smoother and more dynamic than most of the Chinese startups
Click to expand...
You can't seriously be comparing literal slow crawling with Unitree's martial arts or EngineAI's human speed dancing, or compare cartwheel with off the ground side flips. I mean would you compare a 5 year old doing cartwheel with a gymnast?

To be sure they clearly focused a lot on capturing fine motion in their training, but at end of the day Boston Dynamics only switched to reinformcent learning a year ago, and the slow, simple movements they focus on shows it.

Also I think the number of people who claim Unitree or EngineAI's robots are CGI can attest to their smoothness in a blind trial.
 
S

SanWenYu

Captain
Registered Member
SlothmanAllen said:
I feel like all of these humanoid robots are still very early and each on of these companies kind of has a small piece of the overall puzzle that they are specializing in. I think it will be a while before one of these companies is able to bring it all together in one package, or maybe we see various humanoid robots of different configurations in the end. Either way, I think it is very early days for this technology.

Here is cool video so some Figure .02 robots collaborating on a task.

Click to expand...
While collaboration between two humanoids is nice, it is a problem in artificial intelligence. What the robot manufacturers are showing off is the achievements in physical agility of the humanoids.

The ideal humanoid should be agile and intelligent at the same time. To begin with, for most of the simple repetitive tasks that we want humanoids to take, agility is more important than intelligence.
 
P

PeoplesPoster

Junior Member
iewgnem said:
You can't seriously be comparing literal slow crawling with Unitree's martial arts or EngineAI's human speed dancing, or compare cartwheel with off the ground side flips. I mean would you compare a 5 year old doing cartwheel with a gymnast?

To be sure they clearly focused a lot on capturing fine motion in their training, but at end of the day Boston Dynamics only switched to reinformcent learning a year ago, and the slow, simple movements they focus on shows it.

Also I think the number of people who claim Unitree or EngineAI's robots are CGI can attest to their smoothness in a blind trial.
Click to expand...
I actually am seriously making that comparison, because flips and dancing was already done by by BD years ago while the latest videos show, as you admit, much finer control in all aspects of movement. If you think the latest BD video's movements are slow and simple i ask you to take a closer look. unitree has shown great advances in larger movement but it seems the finer controls and smoothness still has work to be done.
 
tankphobia

tankphobia

Senior Member
Registered Member
antwerpery said:

I found the source of the video. It's from a Korean university. Somehow this is 4 year old and somehow they haven't shown off the hand since. Weird. It's so impressive that it should be really big news, especially 4 years ago, you would think that this hand/research team would be famous and everyone would be copying them. But even 4 years later, every humanoid robot has much much worse hands than what this team has made.
Click to expand...
IMO university research is limited to range of motion for robotic actuators rather than intuitively performing a set task, I do not think a university will have the funding/incentive to develop it much further beyond writing a few papers. It is incredibly impressive that this is possible at all, same as all the humanoid robots doing all kinds of complicated stunts.

However, at the end of the day, they are simply following a command created by a programmer, it's not AI in the way that people envision a house robot to automatically do chores or cook. While they're rapidly getting proficiency in performing specific tasks, they're yet to be able to contextualize the surrounding environment and automatically queue up individual actions to accomplish a greater task the same way a human can. I.e to clean this floor I need to pick up this mop bucket, fill it with water, wrung it and mop... etc.

And that's what firms have been doing for the last years. humanoid robots are getting very good at performing individual actions. But the spark to allow them to chain it all together meaningfully is still some time away IMO.
 
You must log in or register to reply here.
Top