Xiaomi’s humanoid assembles EVs autonomously for 3 hours

PLUS: Agibot launches $1k rental + Vision-only robots ace tests


Xiaomi’s humanoid assembles EVs autonomously for 3 hours

Welcome back to your Robot Briefing

Xiaomi just crossed a critical threshold: its humanoid robot worked autonomously on an actual EV production line for three straight hours, hitting a 90.2% success rate on complex assembly tasks. This isn't another controlled demo—it's a live manufacturing environment with real cycle time requirements and variable conditions.

With Tesla, BMW, and Xpeng all racing toward similar deployments, the question facing automation leaders is whether 90% reliability represents the tipping point for investment—or whether the jump to 95%+ success rates will take longer and cost more than current timelines suggest.

In today's Robot update:

Xiaomi's humanoid completes 3-hour autonomous EV assembly run
Vision-only robots clear household task benchmarks years early
Agibot launches $1,000/day robot rental across 17 countries
Military shipbuilder deploys physical AI for autonomous welding
News

Xiaomi's humanoid clocks in: 3 hours of autonomous EV assembly

Snapshot: Xiaomi deployed its humanoid robot on an actual EV production line for three consecutive hours of autonomous work, achieving a 90.2% success rate on complex assembly tasks—a meaningful step beyond controlled demos toward real manufacturing applications.

Breakdown:

The robot handled self-tapping nut installation at Xiaomi's die-casting workshop, completing the task within the production line's 76-second cycle time requirement while coordinating with automated feeding systems and positioning fixtures—demonstrating capability beyond simple pick-and-place operations.
Xiaomi powered the deployment with its proprietary 4.7-billion-parameter vision-language-action model combined with reinforcement learning trained through hundreds of millions of simulated scenarios, enabling the robot to adapt to magnetic interference and variable gripping conditions without extensive manual training data.
The deployment puts Xiaomi in direct competition with Tesla (planning Optimus Gen 3 launch in Q1 2026), BMW (piloting humanoids at Leipzig this summer), and Xpeng (building a mass production base targeting late 2026 output)—CEO Lei Jun previously stated Xiaomi expects large-scale factory deployment within five years.

Takeaway: The 90.2% success rate signals we're still 1-2 years from production-ready humanoid manufacturing, but three-hour autonomous runs represent a substantial leap from scripted demos. Companies evaluating automation roadmaps should track whether success rates cross 95% in the next 12-18 months—that's when the business case shifts from experimental to economically viable.

News

Vision-only robots ace the 'Humanoid Olympics' years ahead of schedule

Snapshot: Physical Intelligence trained robots to complete complex household tasks using only camera inputs, clearing benchmarks experts thought were years away.

Breakdown:

The robots used a vision-only approach with simple pincer grippers to complete 11 of 15 challenges, including tasks like spreading peanut butter and using keys that roboticists believed required expensive force sensors.
Physical Intelligence completed challenges ranging from bronze to gold difficulty in just three months, roughly 15 months faster than the competition designer expected.
The breakthrough suggests robots can learn any task humans can teleoperate, though reliability across different environments with different objects remains the key bottleneck before commercial deployment.

Takeaway: Roboticist Benjie Holson revised his home robot timeline from 15 years away to [six years away](https://www.scientificamerican.com/article/why-humanoid-robots-are-learning-everyday-tasks-faster-than-expected/) based on these results. Companies evaluating automation investments should note that simpler hardware requirements (cameras vs. complex sensors) could accelerate both development timelines and reduce deployment costs significantly.

News

Agibot launches $1,000/day robot rental after topping global shipment charts

Agibot launches $1,000/day robot rental after topping global shipment charts

Image Source: There's A Robot For That

Snapshot: Chinese robotics leader Agibot is launching a Robots-as-a-Service model across Europe and North America, letting companies test humanoid robots without committing to six-figure purchases.

Breakdown:

The rental service starts at $1,000 per day and includes delivery, setup, and operational support, making it viable for short-term events and brand activations but still too expensive for long-term workforce replacement at roughly $360,000 annually.
Agibot offers three robot types across 17 countries : A2 Series humanoids for reception and customer interaction, X2 Series compact units for education and research, and G2 wheeled humanoids optimized for factory assembly and continuous industrial operations.
The company shipped 5,168 units in 2025, leading global humanoid robot shipments and outpacing U. S. competitors like Tesla (150 units) and Figure AI (150 units) by more than 30x.

Takeaway: This rental model creates a low-risk entry point for companies to evaluate humanoid robots in real-world settings, but current pricing positions these as marketing tools rather than operational investments. The massive shipment gap between Chinese manufacturers and U. S. rivals signals that accessible testing models may accelerate adoption timelines faster than anticipated.

News

HII taps 'Physical AI' to automate shipyard welding

Snapshot: Military shipbuilder Huntington Ingalls Industries is partnering with Path Robotics to deploy autonomous welding systems in its shipyards, targeting one of the most labor-intensive bottlenecks in naval construction.

Breakdown:

HII is testing Path Robotics' physical AI-driven welding systems that can autonomously handle complex steel welding tasks without constant human supervision, moving beyond simple robotic arms to systems that adapt to variations in real time.
The partnership addresses a critical pain point in shipbuilding where skilled welders are scarce and welding represents a major time sink in constructing large naval vessels, with defense contractors facing constant pressure on productivity and schedule performance.
HII's approach includes building IP ownership frameworks and workforce training programs rather than just buying equipment, and the technology will integrate with existing programs including the ROMULUS unmanned surface vehicles initiative.

Takeaway: This signals that physical AI has matured enough for highly regulated, mission-critical defense manufacturing environments. Companies in capital-intensive industries with skilled labor constraints should watch whether HII achieves measurable throughput gains, as that would validate the business case for similar automation investments beyond defense.

Other Top Robot Stories

Deloitte unveiled physical AI solutions built with NVIDIA Omniverse for industrial clients, opening a Shanghai center of excellence focused on manufacturing robotics and announcing early results from Horse Powertrain's automotive plant in Spain where AI-powered anomaly detection is reducing equipment downtime and improving quality assurance decision-making.

Automotive analysts questioned whether humanoid form factors offer real advantages for manufacturing, with PwC and Plante Moran experts suggesting that task-specific robots with wheels or alternative designs may deliver superior performance and ROI compared to bipedal humanoids despite the current media attention around Tesla Optimus and competitors.

Intuitive completed its acquisition of da Vinci and Ion surgical robot distribution operations across Italy, Spain, Portugal and Malta for approximately 319 million euros, bringing 250 employees in-house and expanding direct European presence across a region with over 470 installed systems as the surgical robotics market matures beyond single-vendor dominance.

🤖 Your robotics thought for today:

Xiaomi's humanoid hit 90.2% success over three autonomous hours in actual EV assembly while most manufacturers are still waiting for 99% lab performance before they'll even pilot—so does deploying imperfect robots today beat perfecting demos for tomorrow, or are we watching Xiaomi discover that 10% failure rates cost more than the labor they're replacing?

Until tomorrow,
Uli

Xiaomi’s humanoid assembles EVs autonomously for 3 hours

Great! Check your inbox and click the link to confirm.
Please enter a valid email address.