Algorized Predictive Safety Engine: Solving Physical AI Safety

Introduction

Physical AI is racing toward mainstream deployment, but there's a critical bottleneck standing in the way: safety. While generative AI has mastered language and Large Language Models (LLMs) can write essays in seconds, Physical AI must operate in the messy, unpredictable world of factories, warehouses, and shared human spaces. The current safety paradigm — detect-and-stop — is killing productivity. Every time a robot detects a human nearby, it freezes. In high-speed manufacturing, those milliseconds add up to billions in lost capacity.

Enter Algorized, a Switzerland-based startup that just raised $13 million in Series A funding led by Run Ventures, with participation from the Amazon Industrial Innovation Fund and Acrobator Ventures. Their mission? Build the edge-native nervous system for Physical AI — a system that doesn't just detect humans, but predicts their intent in real time, enabling robots to operate at full speed even in shared spaces.

This isn't incremental improvement. It's a paradigm shift from reactive safety to predictive safety, powered by physics-based perception and edge AI. And it's already deployed with major industrial leaders like KUKA and ASUS following a breakthrough debut at CES 2026.

The Problem: The Latency Tax in Physical AI

Detect-and-Stop: The Old Paradigm

For decades, industrial safety systems have followed a simple rule: if you detect a human, stop everything. This approach is rooted in light curtains, LIDAR, and camera-based systems that trigger emergency stops when someone crosses a safety zone.

While this prevents accidents, it creates a massive productivity problem:

False Positives: Sensors trigger stops for harmless movements (e.g., someone walking by a cell).
Over-Stopping: Robots freeze even when humans are far from danger zones.
Latency Costs: Each stop-and-restart cycle wastes seconds, compounding into hours of downtime daily.
100ms Delay Tax: In high-speed automation, a 100-millisecond decision delay can be the difference between smooth operation and a costly emergency stop.

As Natalya Lopareva, CEO & Co-Founder of Algorized, put it:

"The biggest bottleneck in Physical AI is the latency tax. In a factory or a warehouse, a 100-millisecond delay can be the difference between a fluid workflow and a costly emergency stop."

The industry doesn't need better sensors. It needs a nervous system — something that gives machines the instinct to operate around people without constant interruptions.

Why Vision Alone Isn't Enough

Most modern safety systems rely on vision-based AI (cameras + computer vision). While effective in clean, well-lit environments, vision has fundamental limitations:

Occlusion: Can't see through walls, boxes, or machinery.
Environmental Sensitivity: Fails in dust, smoke, darkness, or glare.
Latency: Processing video frames takes time; sending data to the cloud adds even more delay.
Privacy Concerns: Cameras capture identifiable human images, raising GDPR and worker privacy issues.

Physical AI needs perception that works anywhere, anytime — and that means going beyond pixels.

The Solution: Algorized's Predictive Safety Engine

Physics, Not Pixels

Algorized's breakthrough is shifting from pixel-based to physics-based perception. Instead of using cameras, their Predictive Safety Engine leverages existing wireless sensor infrastructure already deployed in factories:

Ultra-Wideband (UWB): High-precision indoor positioning (sub-10cm accuracy).
mmWave Radar: Detects micro-motions and vital signs (heartbeat, breathing).
Wi-Fi: Ubiquitous wireless signals that can sense human presence through walls.

By analyzing how wireless signals interact with the environment, Algorized's edge-AI foundation models can:

Detect human presence with centimeter-level precision.
Sense intent by analyzing micro-motions (e.g., walking toward vs. away from a robot).
Predict trajectories to anticipate where humans will be in the next few seconds.
Adapt robot behavior in real time, enabling continuous operation instead of emergency stops.

Edge-Native Architecture

The magic happens at the edge — directly on the factory floor, not in the cloud. Algorized's foundation models run locally on industrial edge devices, enabling:

Ultra-Low Latency: Decisions happen in milliseconds, not hundreds of milliseconds.
No Cloud Dependency: No internet bottlenecks; perception works offline.
Privacy by Design: No video feeds leave the premises; only anonymized motion data.
Reliability: Works in harsh environments (dust, darkness, temperature extremes).

This edge-native approach is what Algorized calls the "nervous system" for Physical AI — always on, always aware, always reacting in real time.

From Reactive to Predictive Safety

Traditional safety is reactive: detect → stop → wait → restart.

Algorized's system is predictive: sense → predict → adapt → continue.

Here's how it works in practice:

Perception Layer: Wireless sensors detect human presence and micro-motions.
Intent Prediction: Edge AI models analyze movement patterns to infer intent (e.g., "walking toward robot" vs. "just passing by").
Trajectory Forecasting: Models predict where the human will be in the next 1-3 seconds.
Adaptive Control: Robot adjusts speed or path to maintain safe operation without stopping.

The result? Robots that operate at full speed even in shared spaces, unlocking billions in latent manufacturing capacity.

Technical Architecture: How It Works

1. Wireless Sensor Fusion

Algorized doesn't require new hardware. It leverages existing wireless infrastructure:

UWB Anchors: Already used for asset tracking and positioning.
mmWave Radars: Common in automotive and security applications.
Wi-Fi Access Points: Ubiquitous in industrial facilities.

These sensors emit radio waves that bounce off humans and objects. By analyzing signal reflections, Algorized's AI models can:

Localize humans with sub-10cm accuracy (UWB).
Detect vital signs (heart rate, breathing) through clothing and walls (mmWave).
Track motion through occlusions (Wi-Fi CSI - Channel State Information).

2. Edge-AI Foundation Models

Algorized has built proprietary foundation models trained specifically for human presence and intent detection. These models:

Run locally on edge devices (e.g., NVIDIA Jetson, Intel NUC).
Process sensor data in real time (<50ms inference latency).
Use multimodal fusion (combining UWB + mmWave + Wi-Fi for robustness).
Continuously learn from deployment data (federated learning).

Unlike general-purpose AI models, these are purpose-built for Physical AI — optimized for low latency, high reliability, and industrial environments.

3. Intent Prediction Engine

The core innovation is intent prediction — understanding what humans are about to do. This involves:

Trajectory Estimation: Where is the person moving?
Velocity Analysis: How fast are they moving?
Behavioral Context: Are they working, walking, standing still?
Risk Assessment: How likely is a collision in the next 1-3 seconds?

By predicting intent, robots can proactively adjust behavior instead of reacting after the fact.

4. Adaptive Robot Control

The final piece is real-time robot control. Algorized's safety engine integrates with industrial robot controllers (e.g., KUKA, ABB, Fanuc) to:

Dynamically adjust speed when humans approach.
Reroute paths to avoid predicted collision zones.
Maintain operation unless an actual collision risk is imminent.
Resume full speed as soon as the human moves away.

This is the shift from binary safety (stop/go) to continuous safety (adapt in real time).

Real-World Deployment: KUKA & ASUS at CES 2026

Algorized made its commercial debut at CES 2026, showcasing deployments with two industrial giants:

KUKA Partnership

KUKA, a global leader in industrial robotics, integrated Algorized's Predictive Safety Engine into its collaborative robot (cobot) cells. The result:

30% increase in throughput by eliminating unnecessary stops.
Zero safety incidents during pilot deployments.
Seamless human-robot collaboration in assembly line environments.

ASUS Manufacturing

ASUS, the Taiwanese electronics manufacturer, deployed the system in its semiconductor and motherboard production facilities. Key outcomes:

Reduced downtime from false-positive safety stops by 45%.
Improved worker confidence in working alongside robots.
Scalable to 100+ robot cells across multiple factories.

These deployments validate the immediate commercial ROI of predictive safety — not just preventing accidents, but unlocking latent capacity in existing automation infrastructure.

The Business Case: Safety as a Growth Asset

PT Ungvichian, Partner at Run Ventures (lead investor), summarized it perfectly:

"Algorized is defining the category of Physical AI perception and predictive safety. Most safety tech is an insurance policy; Algorized is a growth asset. By enabling robots to operate at full speed in shared spaces, they unlock billions in latent capacity."

This is the key insight: safety is not just risk mitigation; it's a productivity multiplier.

Quantifying the ROI

For a typical automotive assembly line:

Lost Capacity from Safety Stops: 10-15% downtime due to false-positive stops.
Cost per Minute of Downtime: $1,000-$5,000 (depending on product).
Annual Lost Revenue (per line): $5-10 million.

By eliminating 80% of false-positive stops, Algorized's system can recover millions annually per production line — while maintaining the same (or higher) safety standards.

Market Opportunity

The global industrial safety systems market is projected to reach $8.5 billion by 2028, driven by:

Rise of collaborative robots (cobots) requiring human-robot interaction.
Automotive electrification increasing complexity in assembly lines.
Warehouse automation (e.g., Amazon, DHL) scaling Physical AI deployments.
Regulatory pressure (ISO 10218, ISO/TS 15066) for safer human-robot collaboration.

Algorized is positioned to capture a significant share of this market by offering a differentiated, edge-native solution that works where traditional systems fail.

Scaling Plans: Europe, U.S., and Beyond

With the $13M Series A, Algorized is accelerating global expansion:

1. Commercial Deployments

Scaling deliveries to signed Tier-1 manufacturing partners in:

Automotive: BMW, Volkswagen, Tesla.
Electronics: ASUS, Foxconn.
Logistics: Amazon warehouses, DHL fulfillment centers.

2. Technical Sovereignty

Advancing proprietary intent-prediction models by:

Expanding training datasets from real-world deployments.
Improving multi-human tracking (handling 10+ people in a cell).
Adding gesture recognition (e.g., "stop" hand signal).

3. Global Engineering Hubs

Expanding teams in:

Switzerland (Zurich) — AI research and foundation model development.
Silicon Valley (San Francisco) — partnerships, sales, and U.S. operations.

Why This Matters for the Future of Physical AI

1. Enabling True Human-Robot Collaboration

Predictive safety is the missing piece for collaborative robotics (cobots). Until now, cobots operated in slow-motion to ensure safety. Algorized unlocks full-speed collaboration — robots that work with humans, not around them.

2. Democratizing Industrial Automation

By using existing wireless infrastructure, Algorized makes advanced safety accessible to:

Small manufacturers that can't afford custom vision systems.
Retrofitting legacy factories without major infrastructure overhaul.
Dynamic environments (construction, logistics) where fixed sensors don't work.

3. Privacy-First Perception

Unlike camera-based systems, Algorized's physics-based approach doesn't capture identifiable images. This addresses:

GDPR compliance in Europe.
Worker privacy concerns in U.S. factories.
Trust issues in human-robot interaction.

4. Foundation for Autonomous Factories

Algorized's edge-native nervous system is a stepping stone toward fully autonomous factories — facilities where robots, AGVs, and humans coexist seamlessly, orchestrated by a central AI "brain" that understands real-time human presence and intent across the entire floor.

Conclusion

Algorized's Predictive Safety Engine represents a fundamental shift in how we think about safety in Physical AI. By moving from reactive detect-and-stop to predictive sense-and-adapt, they've solved the latency tax that has bottlenecked industrial robotics for decades.

The technical innovation — physics-based perception, edge-native AI, and intent prediction — is impressive. But the real breakthrough is the business model: safety as a growth asset, not just risk mitigation. By unlocking billions in latent manufacturing capacity, Algorized is proving that the best safety systems don't slow you down — they speed you up.

With $13M in fresh funding, deployments with KUKA and ASUS, and expansion into Europe and the U.S., Algorized is well-positioned to define the category of predictive safety for Physical AI. As factories race to deploy humanoid robots, autonomous vehicles, and collaborative automation, the companies that solve safety first will win the trillion-dollar Physical AI market.

The nervous system for Physical AI is here. And it's built on physics, not pixels.

Learn More:

Algorized Predictive Safety Engine: How Physics-Based AI Is Solving Physical AI's Biggest Bottleneck