00:00Our goal is to create the most human-like robot possible.
00:10This vision drives every team at XP'ing Robotics,
00:13uniting our efforts in industrial design,
00:15mechanical engineering, motion control, and beyond.
00:19We've always wanted our robot to feel familiar.
00:23It moves with a quiet kind of grace,
00:25soft arms, warm gestures, and hugs
00:29To achieve that, we developed a general-purpose humanoid design framework.
00:35It is not just an aesthetic guideline.
00:38It is a body logic that keeps the robot balanced, expressive, and coherent
00:44across different sizes and use cases.
00:47Based on this framework, we created the new generation Iron,
00:51a general-purpose bionic humanoid built layer by layer
00:56from a compact skeleton to its muscle-like lattice structures
01:00to the soft outer skin that makes it feel almost human to the touch.
01:05Every part of the human body is really a masterpiece.
01:10The way our ligaments, muscles, and bones work together
01:13gives us an incredible range of motion.
01:17To make our robot as human-like as possible,
01:20we carried out a deep biomimetic study based on human anatomy,
01:26and then followed it up with solid engineering optimization.
01:31When we were developing its movements,
01:34we realized just how important the waist degrees of freedom are.
01:39Instead of using simple rotational joints,
01:42we designed a structure that's more like a real human spine.
01:48We also tried adding even more DOF to see if we could boost performance,
01:54but of course, that made the control systems more complex.
01:59Thanks to all this work,
02:00our robot can now do genuinely human-like actions,
02:04things like shrugging or hugging,
02:07and even simple motions such as nodding, dancing, or walking,
02:13now look much more natural and fluid.
02:17It's actually very similar for robots.
02:20They also need highly efficient learning methods to perform at a top level.
02:24We have basically rebuilt our reinforcement learning algorithms from scratch,
02:29training motion controllers that are extremely robust.
02:31Because of that, our robot can keep a smooth human-like motion,
02:36even when its muscle structure or skin materials change.
02:40The problem was the unique properties of the lattice material
02:43made it extremely hard to simulate its dynamic effects efficiently.
02:48In simple terms,
02:50no one could really tell how the muscles would affect the system's dynamics.
02:54To outcome that,
02:56we have collected a massive amount of dynamic data
02:59and developed our own proprietary algorithms for system identification.
03:05Especially tailored to the muscles' lattice,
03:08then we used the computational methods to optimize the structure.
03:12On top of that,
03:13we precisely calibrated the robot's model parameters,
03:17which finally made efficient simulation of the lattice structure possible.
03:21with the coordinated cooperation of hardware configuration,
03:26software control,
03:27and appearance design,
03:29with flexible waist twisting,
03:31and natural shoulder shrugging.
03:33Our new generation robot showed an elegant catwalk at the technology day.
03:38Every step forward of XPeng means that we are getting a little closer
03:42to our vision of perfect collaboration
03:44between machine intelligence and human wisdom.
03:50For more videos,
03:52just subscribe.
03:53aking environmental disparities,
03:53green lake etc.
03:54For more videos,
03:58we are in an overall space,
03:58support us at the laquelle we will be perfectly
04:16compared withpresa.
Kommentare