# computer vision

Carbon Robotics says one plant model can retune laser weeders in minutes

Carbon Robotics says it replaced crop-specific vision models with a single large plant model.
Farmers can label a small number of field images in an iPad app to personalize the system on site.
The company positions the change as a way to speed deployment of autonomous laser weeding across crops and regions.

DT Editorial Team·Aug 28, 2026·via therobotreport.com

‘Count Anything’ Aims to Turn Object Counting Into a General AI Skill

The model combines box-based and point-based counting methods.
It is built on Meta’s SAM3 with added counting adapters.
Researchers report strong results but note weaknesses in ambiguous and dense scenes.

DT Editorial Team·Jun 14, 2026·via the-decoder.com

US makes 'artificial eyes' to help robot, self-driving cars see better

Innovation

Human-Eye-Inspired 'Artificial Eyes' Could Prevent Self-Driving Cars from Going Blind

Penn State engineers create an artificial eye inspired by the human eye.
Device prevents self-driving cars from losing vision in challenging conditions.
Curved photodetector array and adaptive lens mimic human retina and iris.
Technology applicable to robotics, manufacturing, and space exploration.

DT Editorial Team·Jun 9, 2026·via interestingengineering.com

Robots still fail outside the demo because vision breaks first

The essay says deployment failures often begin in sensing, not planning.
2D vision remains useful but is limited for many real-world robotics tasks.
Reliable uncertainty and calibration are presented as core requirements.

DT Editorial Team·May 28, 2026·via therobotreport.com

China is upgrading older surveillance cameras with embedded AI systems that can analyze behavior, search video with tex

Manufacturers are equipping older surveillance networks with AI vision and language-model tools.
The systems can flag behavior such as crowds, erratic driving, and unauthorized access.
The shift appears to move China’s surveillance model toward broader automated behavioral monitoring.

DT Editorial Team·May 28, 2026·via the-decoder.com

How humanoids learn to read the room - The Robot Report

Humanoid Robots Are Being Built Around a Harder Problem Than Motion: Human Proximity

The key humanoid challenge is operating safely around unpredictable humans.
Vision and depth sensing matter, but latency and data transport are central constraints.
The article points to automotive-style sensor architectures as a model for robotics.

DT Editorial Team·May 27, 2026·via therobotreport.com

Yes, you need a smart bird feeder in your life - and this one's on sale for Memorial Day

News

Consumer Tech Turns Backyard Birding Into an AI Product Category

Birdfy’s smart feeder uses a camera and AI features to capture bird visits and send alerts.
The product was featured in a Memorial Day sale at 35% off.
The device reflects a broader trend of AI moving into hobby-focused consumer hardware.

DT Editorial Team·May 23, 2026·via zdnet.com

Pentagon Pushes AI Targeting to Speed Drone Defense From Turrets to Rifles

The Pentagon wants AI-assisted target recognition to speed drone defense and reduce false identification.
Initial work focuses on CROWS remote weapon stations, with later phases expanding to maritime platforms and small arms.
The effort highlights how central low-cost drone threats have become to modern tactical planning.

DT Editorial Team·May 8, 2026·via defensenews.com

Anthropic Pushes Claude Opus 4.7 Further Into Coding While Deliberately Limiting Cyber Use

Anthropic says Claude Opus 4.7 scored 64.3% on SWE-bench Pro, up from 53.4% for Opus 4.6.
The model now handles much higher image resolution and showed large gains on document reasoning.
Anthropic says it deliberately reduced risky cybersecurity capabilities during training.

DT Editorial Team·Apr 16, 2026·via the-decoder.com

AI Models Still Prefer Guessing Over Asking for Help, Benchmark Finds

ProactiveBench tests whether multimodal AI models ask for missing visual information.
Across 22 models, average performance fell from 79.8 percent in standard settings to 17.5 percent on the proactive benchmark.
The findings suggest current systems are better at guessing than recognizing when they need help.

DT Editorial Team·Apr 12, 2026·via the-decoder.com

Netflix open-sources VOID, an AI framework that erases video objects and rewrites the physics they left behind

Netflix’s VOID Pushes Video AI Beyond Object Removal and Into Cause-and-Effect Editing

Netflix has open-sourced VOID, a framework for removing objects from video scenes.
The system also adjusts downstream physical effects such as collisions left by removed objects.
VOID combines CogVideoX, Gemini 3 Pro, SAM2, synthetic training data, and optional optical-flow correction.

DT Editorial Team·Apr 5, 2026·via the-decoder.com