From Cristiano Ronaldo’s personal mid-air spin party to Kobe Bryant’s fadeaway leap shot, the collaboration between Carnegie Mellon University and NVIDIA has resulted in a new training technique that enables human robots to perform complex athletic movements with unmatched agility.

Aligning Simulation and Real Physics ( ASAP ) is a key step in the development of a framework that allows humanoid robots to perform high-level athletic movements that were previously deemed too difficult for machines.

” Humanoid drones hold the potential for unparalleled flexibility for performing human-like, whole-body abilities”, the scientists noted in their report. ” However, the interactions mismatch between model and the real world continues to present a significant problem in terms of achieving agile and coordinated whole-body movements.”

ASAP tackles this obstacle through a two-stage approach.

Second, it pre-trains action tracking policies—the analytic rules that control the tracking—in simulation using animal movement data. Then it applies these guidelines in the real world to gather data that helps bridge the gap between replicated and true physics.

The result is a humanoid robot capable of s://x.com/minchoi/status/1886796078558634156″ target=”_blank” rel=”nofollow external noopener” class=”sc-adb616fe-0 bJsyml”>replicating signature moves from sports legends, including Cristiano Ronaldo’s iconic” Siu” celebration ( involving a 180-degree mid-air rotation ), LeBron James’s” Silencer” celebration ( featuring precise single-leg balancing ) and Kobe Bryant’s fadeaway jump shot ( which involves jumping and landing in one foot ).

Beyond these extravagant athletic moves, the machine also demonstrated amazing feats like side jumps and onward jumps of more than a meter.

At first glance, the robots may also look stupid, but this time, it’s mainly due to technology limitations, as they have means less expression than a man.

Due to the “delta action design,” a revision system that makes up for the distinctions between modeled and actual physics, they are more dexterous than other robots. Essentially, the terminal action model serves as the dynamics gap’s remaining correction term.

Using this method, experts reduced tracking problems by up to 52.7 % compared to previous methods, enabling drones to perform complex movements that were formerly impossible.

” Our approach drastically improves agility and whole-body coordination across several powerful motions”, noted the experts, who demonstrated the system’s performance is “paving the means for versatile humanoid drones in real-world software”.

One of the most frequent problems in robotics has been developing drones with this level of skill.

” For years, we have envisioned human computers achieving or even achieving human-level agility. However, most earlier work has largely focused on movement, treating the arms as a means of mobility”. the experts wrote.

ASAP, on the other hand, resembles the human body in pre-train and is able to adjust its expertise to real-world situations after learning about it in models.

This means, the creature’s legs behave of human legs, being used for motion, balance, counterpoint, expression, and more.

That is a little harder than it seems to be. When we perform athletic—and also basic—moves, we’re really orchestrating many simple changes in real-time, balancing several forces while compensating for changes in speed and location.

It has been extremely challenging to get robots to replicate this.

Don’t believe us? Try playing the QWOP, a game where you must control four articulations to a sprinter. Think about how challenging it would be to manage the 21 fundamental articulations that ASAP handles once you have worked hard to master that game while also considering that the human body has over 300 different joints.

In recent years, the field of humanoid robotics has grown in popularity, with more companies and universities putting more money into R&D.

Tesla’s Optimus project, Figure AI’s recent humanoid robot announcement, and Boston Dynamics ‘ Atlas have all highlighted growing commercial interest in humanoid robots.

The University of Bristol and Stanford have also developed their own methods to teach models how to be more agile and develop their dexterity in the academic area.

The team is focused on developing more quickly.

According to them,” Future directions could concentrate on developing damage-aware policy architectures to mitigate hardware risks,” referring to how some models broke while performing complex moves.

Additionally, they want to study how to improve their adaptation methods to achieve higher efficiency by using markerless pose estimation or onboard sensor fusion to reduce MoCap systems ‘ reliance. How long until an all-robot World Cup is held?

edited by Josh Quittner and Sebastian Sinclair

Generally Intelligent Newsletter

A generative AI model’s generative AI model, Gen, tells a weekly AI journey.

Share This Story, Choose Your Platform!