As makers race to deploy humanoid robots in factories, warehouses and eventually homes, a fundamental question has resurfaced: how do you design one so it does not hurt a person? According to The Wall Street Journal, a series of recently circulated "mishap" videos underscores the problem. The article notes that AI-driven humanoids carry an unpredictability that traditional industrial robots do not.
Humanoid Robots · Safety by Design
The Quest to Make Humanoid Robots Safe Enough for Humans
As machines grow heavier and act on their own AI-driven judgment, makers face one hard problem: guaranteeing a robot won't hurt a person. Viral mishaps have renewed the urgency — and reshaped how the machines are being designed.
1B
humanoid units forecast worldwide by 2050
$7.5T
projected market size by 2050
2028
target for first ISO stability standards for humanoids
Two ways a machine decides
The core of the problem: humanoids do not follow fixed rules — they judge situations on the fly, which makes their behavior harder to guarantee.
DETERMINISTIC
Traditional industrial robots
Welders, palletizers — follow set rules, same action every time. Predictable.
PROBABILISTIC
AI-driven humanoids
Judge dynamically via sensors, cameras and AI. Flexible — but unpredictable.
Heavier machines, bigger risk
As humanoids scale up, a power loss can mean toppling or crushing. Weight is now a safety factor.
~80 kg
Neura 4NE1
~90+ kg
Heaviest models
The heavier the machine, the greater the crushing risk if it loses power and falls.
Layered safety — from E-stop to chip level
E-STOP
Emergency-stop buttons for instant human override.
→
SENSOR AI
Nvidia Blackwell chips analyze data and halt on danger.
→
SAFE FALL
Design to collapse onto itself when a fault is detected.
On the floor today: Agility's Digit runs inside a Plexiglas cage at a Schaeffler plant; wheeled models keep a low center of gravity to cut fall risk. Full human-robot collaboration stays limited until standards mature.
The bottleneck
Billions are flowing in and factories welcome robots for dangerous work — but expanded deployment will move only in step with the buildout of safety standards, expected by mid-2028.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…