Carbon Robotics debuts 150M-image 'Large Plant Model'

PLUS: LimX raises $200M, Unitree G1 braves -47°C, Themis works while walking


Carbon Robotics debuts 150M-image 'Large Plant Model'

Welcome back to your Robot Briefing

Carbon Robotics just eliminated the 24-hour retraining delay that's plagued agricultural AI, unveiling a Large Plant Model trained on 150 million images that lets farmers identify new weeds and adjust robot behavior in minutes through an iPad interface. This shift from brittle vision systems to foundation models means deployment friction is dropping faster than most operations leaders expected.

If specialty crop automation is reaching this maturity level, how much runway do labor-intensive industries really have before similar breakthroughs force their hand on robotics investments?

In today's Robot update:

Carbon Robotics debuts 150M-image plant foundation model
LimX raises $200M, demos 18-humanoid coordinated swarm
Westwood's Themis manipulates objects while walking
Unitree G1 walks 130,000 steps in -47°C conditions
News

Agriculture's 'ChatGPT Moment': Carbon Robotics Unveils Large Plant Model

Agriculture's 'ChatGPT Moment': Carbon Robotics Unveils Large Plant Model

Image Source: There's A Robot For That

Snapshot: Carbon Robotics launched a Large Plant Model that lets its laser-equipped weed-killing robots identify and target new plant species instantly, eliminating the 24-hour retraining cycle that previously delayed operations whenever unfamiliar weeds appeared in fields.

Breakdown:

The model trained on 150 million plant images from 100+ farms across 15 countries, allowing the LaserWeeder fleet to recognize any weed species on sight without manual retraining or coding updates.
Farmers now adapt the system through a feature called Plant Profiles by selecting 2-3 images on an iPad interface, enabling the robots to adjust behavior in minutes rather than the weeks or months required by traditional agricultural AI systems.
CEO Paul Mikesell, who previously built neural networks at Uber and Meta's Oculus division, told TechCrunch that farmers can now say in real-time "this is a new weed, kill this" without waiting for engineering support.

Takeaway: Agricultural robotics is shifting from brittle computer vision systems that break when conditions change to foundation models that generalize across environments. For operations leaders evaluating automation ROI, this development signals that deployment friction and downtime costs are dropping faster than expected in specialty crop sectors where labor and herbicide costs create immediate pressure to adopt.

News

LimX Dynamics Raises $200M, Deploys Swarm of 18 Humanoids

Snapshot: Chinese robotics firm LimX Dynamics secured $200 million in Series B funding and demonstrated 18 humanoid robots standing up and moving in coordination without individual human commands—signaling that the industry is shifting from proving robots work to proving they can deploy at scale.

Breakdown:

The funding round included strategic investors like JD.com, SAIC Motor-backed Shangqi Capital, and NIO Capital, with LimX planning to accelerate hardware manufacturing, motion control models, and its proprietary operating system for real-world deployment.
LimX's modular TRON 2 platform lets companies reconfigure a single robot across multiple forms and tasks without custom hardware redesigns, directly addressing the cost and complexity that has kept robotics confined to specialized applications.
The 18-robot demonstration showed autonomous fleet coordination where machines powered on together, maintained spacing, and moved as a unit—the kind of scalable deployment that warehouses and factories need but rarely see outside controlled labs.

Takeaway: The combination of substantial institutional backing and demonstrated fleet autonomy suggests humanoid robotics is transitioning from science projects to industrial tools faster than many mid-sized companies realize. Leaders who assume "humanoids are still 5 years away" may want to revisit that timeline—the gap between pilot programs and practical deployment is closing quickly.

News

Westwood's New Humanoid Can Finally Work While Walking

Snapshot: Westwood Robotics launched Themis Gen 2.5, a humanoid that manipulates objects while walking rather than stopping to work—a capability that directly impacts productivity in logistics and manufacturing environments.

Breakdown:

The robot uses a whole-body loco-manipulation controller that maintains balance and precision during simultaneous walking and object handling, eliminating the stop-start inefficiency common in current humanoid deployments.
Hardware upgrades include 40% greater impact resistance, 7-degree-of-freedom arms with 5kg payload capacity per arm, and new hip actuators that deliver 120% more torque while generating 80% less heat.
The system runs on AOS (AI-Augmented Humanoid OS), which integrates perception, navigation with multi-layer mapping, and an object-centric vision-action model for compute-efficient task planning.

Takeaway: Continuous manipulation while moving addresses one of the biggest productivity bottlenecks in humanoid robotics—time lost stopping to work. This positions Themis as a contender for environments where mobile picking, sorting, or assembly operations need to maintain flow rather than operate from fixed stations.

News

Unitree's G1 Robot Conquers -47°C Snowfield in Endurance Test

Snapshot: Unitree's G1 humanoid robot walked over 130,000 steps across a snowfield at -47.4°C in China's Xinjiang region, marking the first autonomous walking demonstration by a humanoid robot in extreme cold conditions.

Breakdown:

The robot autonomously traced a 186-meter Winter Olympics emblem across snow and ice while maintaining balance and navigation in conditions that would disable most electronic systems, demonstrating hardware reliability beyond controlled lab environments.
Unitree shipped over 5,500 humanoid units in 2025 and prices the G1 starting at $14,240 , positioning it among the most accessible humanoid platforms while now proving viability for harsh-environment applications like mining, energy infrastructure, and cold-chain logistics.
The demonstration relied on China's Beidou satellite system for centimeter-level navigation and the robot's adaptive path planning, showing that supporting infrastructure for autonomous operation in challenging conditions is maturing faster than many expected.

Takeaway: Extreme weather reliability has been a critical barrier preventing humanoid deployment in outdoor industrial settings like construction sites, warehouses, and energy facilities. Unitree's test suggests this technical hurdle is being cleared sooner than anticipated, potentially accelerating the timeline for companies evaluating humanoids for year-round outdoor operations.

Other Top Robot Stories

Fraunhofer developed AI-powered autonomous drones and ground robots that detect radioactive, chemical, and biological hazards in environments too dangerous for human access, with field trials showing the systems can pinpoint contamination sources within meters in minutes rather than the days required by traditional handheld detection methods.

China accelerated hybrid crop development by 400% using robotic breeding systems that autonomously pollinate and harvest "robot-friendly" crops, while deploying autonomous robots across farms to handle pest detection, livestock monitoring, and even robotic fish that guide schools to designated feeding and harvesting areas without disturbing aquatic environments.

Tsubame operated its Archax mech for three years with virtually no maintenance while maintaining full functionality across normal operations, transformation sequences, and locomotion—a reliability milestone that demonstrates industrial robots can sustain complex multi-mode performance without constant technical support.

GITAI assembled a 5-meter communication tower using three inchworm robotic arms coordinating with a lunar rover in vacuum chamber testing, demonstrating that multi-robot construction teams can handle complex structural assembly tasks in extreme environments where human labor is impossible or prohibitively expensive.

🤖 Your robotics thought for today:
A farmer updates her weeding robot's vision in 3 minutes on an iPad. No engineer. No vendor call.
Meanwhile, a $500M distribution center is still waiting 6 weeks for the same thing.
The foundation model moment hit agriculture first. Warehouses are next — whether they're ready or not.

Until tomorrow,
Uli

Carbon Robotics debuts 150M-image 'Large Plant Model'

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