惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Martin Fowler
Martin Fowler
Blog — PlanetScale
Blog — PlanetScale
Vercel News
Vercel News
L
LangChain Blog
Google DeepMind News
Google DeepMind News
H
Hackread – Cybersecurity News, Data Breaches, AI and More
F
Fortinet All Blogs
The GitHub Blog
The GitHub Blog
Recent Announcements
Recent Announcements
D
DataBreaches.Net
云风的 BLOG
云风的 BLOG
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
A
About on SuperTechFans
博客园_首页
N
Netflix TechBlog - Medium
Y
Y Combinator Blog
Hugging Face - Blog
Hugging Face - Blog
Last Week in AI
Last Week in AI
酷 壳 – CoolShell
酷 壳 – CoolShell
B
Blog
Apple Machine Learning Research
Apple Machine Learning Research
罗磊的独立博客
美团技术团队
V
V2EX

Interesting Engineering

US firm to scale laser-based nuclear fusion ‘breakthrough’ with new partnership Military Archives - Interesting Engineering World’s first non-nuclear lead-cooled reactor to generate electricity begins installation US scientists devise new process to turn sewage sludge into 99% pure natural gas US firm unveils submarine-hunting drone with 9,200-mile-range, 35 mph top speed Military Archives - Interesting Engineering Supercomputer finds lithium-titanium tweak to boost sodium-ion batteries for grids Lockheed Martin demonstrates vertical launch missile system for mobile drone defense China’s 1116 MWe Taipingling Unit 1 reactor goes online, set to generate 9bn kWh yearly ChatGPT Images 2.0 update combines reasoning, research, and design with 2K output US Navy tests plug-and-play laser system on USS Bush carrier, downs drones at sea China’s CATL reveals 621-mile EV battery, under-7-minute charging to challenge BYD US uses world’s first exascale supercomputer to model supernovae, fusion reactors AI and Robotics Archives - Interesting Engineering First-in-human study confirms safety of graphene-based brain interface Tesla’s Optimus humanoid robot greets runners, poses for photos at Boston Marathon Interlocking materials offer high strength and flexibility for robotics, infrastructure US redeploys 100,000-ton nuclear-powered aircraft carrier in Red Sea after repairs US scientists unveil concept for ‘world’s first neutrino laser’ to unlock breakthroughs New military tech can maintain communication in contested electronic warfare environments Got a dark personality? Psychologists can help you choose your career wisely Humidity boosts performance of 3D-printed nanogenerator instead of degrading it China demonstrates microwave beam that recharges drones in flight, continues power delivery Scientists run compact free-electron laser for eight hours, cracks FEL stability problem China’s PLA considers to use minelaying underwater drones to enforce Taiwan blockade: Report 1-ton sharks may struggle for survival in waters exceeding 62.6°F, study suggests US firm’s thorium nuclear fuel bundles move to manufacturing for commercial reactors Tesla hits 0% charge in remote Chilean desert as YouTuber uses hood-mounted solar Humanoid robot surpasses human world record in Beijing half-marathon, clocking 50:26 mins New method extracts maximum work from unknown quantum states using symmetry tricks
Watch Boston Dynamics’ Atlas humanoid robot amazes with f...
Jijo Malayil · 2026-05-06 · via Interesting Engineering

Each move shows precise coordination, with real-time joint adjustments at hips and ankles to maintain balance and absorb impact.

Boston Dynamics says Atlas learns via simulations and tests, enabling self-positioning without external cues.

Boston Dynamics says Atlas learns via simulations and tests, enabling self-positioning without external cues.Boston Dynamics/YouTube

Engineers at Boston Dynamics have released new test footage of the Atlas humanoid robot, showcasing its advanced balance and mobility.

The robot is seen shifting from standing on two feet into unusual, off-center positions, repeatedly challenging its stability with each movement.

Atlas shifts onto one leg with coordinated arm balance, then places both hands down and smoothly transitions into a controlled handstand.

In January 2025, Atlas was showcased performing complex movements, showing simulation-trained control can transfer to real-world robots and future factory use.

Dynamic motion leap

New footage from Boston Dynamics showcases the Atlas humanoid robot executing a sequence of advanced balance and gymnastics maneuvers that underscore the system’s evolving control capabilities.

The robot transitions from a stable standing posture into a fully inverted handstand, supporting its entire body weight on its palms despite the limited contact area. While inverted, it rotates its legs forward in a controlled 180-degree arc, enabled by highly flexible shoulder joints that allow smooth forward and backward motion, reports Techeblog.

Atlas then reverses direction, extending its legs backward into a sustained L-sit position, holding the posture steadily for several seconds. The routine continues with fluid recovery motions as the robot returns upright and performs additional dynamic movements with minimal instability.

Each action demonstrates precise coordination across its upper body, core, and limbs, with real-time adjustments through joints such as the hips and ankles to maintain balance and absorb impact forces. Reports suggest that these demonstrations highlight the sophistication of Atlas’s whole-body control system, pointing toward practical applications where robots must operate in complex, constrained environments.

Atlas humanoid robot nears production

In the last few months, Boston Dynamics has been approaching the final phase of testing for the Atlas humanoid research platform, using high-intensity acrobatic trials to validate its underlying control architecture before transitioning to a production-ready system.

In collaboration with the Robotics & AI Institute, the robot has been subjected to complex full-body motions—including dynamic locomotion, aerial maneuvers, and recovery from instability—to stress-test its software stack.

At the core of these demonstrations is a unified, simulation-trained control framework based on reinforcement learning. Unlike traditional robotics pipelines that separate locomotion and manipulation, Atlas uses a whole-body learning approach that coordinates arms, legs, torso, and joints as a single system.

According to the firm, it enables behaviors such as cartwheels, backflips, and rapid balance correction to emerge from the same policy that governs walking. The system is designed for “zero-shot” transfer, meaning policies trained in simulation can be deployed directly onto hardware without task-specific tuning.

The control stack continuously adjusts joint torques and contact forces in real time, allowing Atlas to absorb impacts, recover from missteps, and maintain stability during unpredictable interactions. These capabilities are further enhanced by full-body contact strategies, where limbs dynamically assist balance and motion rather than operating independently.

While the research version of Atlas performs acrobatics for demonstrations, the production model is being readied for industrial use. The enterprise-focused Atlas is built for large-scale manufacturing, featuring 56 degrees of freedom and a four-digit, tactile-sensing gripper.

Hyundai Motor Group, of which Boston Dynamics is a subsidiary of, has confirmed deployment at the Hyundai Motor Group Metaplant America by 2028, where initial duties will involve parts sequencing before expanding into full component assembly by 2030.

The Blueprint

Get the latest in engineering, tech, space & science - delivered daily to your inbox.

Jijo is an automotive and business journalist based in India. Armed with a BA in History (Honors) from St. Stephen's College, Delhi University, and a PG diploma in Journalism from the Indian Institute of Mass Communication, Delhi, he has worked for news agencies, national newspapers, and automotive magazines. In his spare time, he likes to go off-roading, engage in political discourse, travel, and teach languages.