Brainy humanoid robot masters package sorting at lightning 4-second pace

🧠 Weekly Brief

Brainy humanoid robot masters package sorting at lightning 4-second pace

Summary: Figure AI's Figure 02 robot achieved a breakthrough in warehouse automation, processing packages in 4.05 seconds each—a 20% speed improvement—during hour-long continuous operation. The robot now handles diverse packaging including deformable poly bags and flat envelopes, achieving 95% barcode scanning success through enhanced memory and force feedback.

3 Key Takeaways:

• Training data multiplication pays off: Six-fold increase in training demonstrations (from 10 to 60 hours) combined with new visual memory and force feedback modules enabled the robot to handle complex, non-rigid packaging that would stump earlier versions

• Adaptive behaviors emerge naturally: The robot learned nuanced techniques like "patting down bubbled plastic so a wrinkled label will scan cleanly" and flicking soft bags to flip them—behaviors that emerged from end-to-end learning rather than explicit programming

• Generalization hints at scalability: With minimal additional training, the same system learned to recognize human hand signals for package handoffs, suggesting one learning pipeline could eventually cover "dozens of warehouse micro-jobs, from kitting bins to palletizing"

Source

Hexagon Taps NVIDIA Robotics and AI Software to Build and Deploy AEON, a New Humanoid

Summary: Hexagon unveiled AEON, a humanoid robot built with NVIDIA's three-computer robotics platform to address 50 million unfilled positions globally. Designed for reality capture, manipulation, and inspection across automotive, aerospace, and manufacturing sectors, AEON creates digital twins by scanning assets and uploading data to cloud platforms.

3 Key Takeaways:

• Simulation drastically accelerates development: AEON mastered core locomotion skills in just 2-3 weeks using NVIDIA Isaac Sim and Isaac Lab, compared to the typical 5-6 months required for real-world training—demonstrating the power of simulation-first robotics development

• Digital twin automation breakthrough: AEON autonomously scans everything from precision parts to full assembly lines, with captured data flowing directly into Hexagon Reality Cloud Studio via NVIDIA Omniverse, potentially making digital twins "a mainstream tool for collaboration and innovation"

• Full-stack AI integration realized: The robot leverages NVIDIA's complete three-computer system—AI supercomputers for training, Omniverse for simulation testing, and IGX Thor for deployment—representing a mature end-to-end physical AI pipeline ready for industrial use

Source

World's first flying humanoid robot uses AI and jet thrust to hover in test flight

Summary (50 words): Italy's iRonCub3 became the world's first jet-powered humanoid robot to achieve controlled flight, hovering 50 cm with four jet engines generating over 1000 newtons of thrust. The 70kg robot required a titanium spine to handle 800°C exhaust temperatures and AI-powered control systems for stability.

3 Key Takeaways:

• Engineering breakthrough for asymmetric flight: Unlike symmetrical drones, iRonCub3's elongated body with movable limbs creates shifting aerodynamics and dynamic center of mass, requiring entirely new flight control models and real-time aerodynamic estimators developed through collaboration with Polytechnic of Milan and Stanford University

• Co-design methodology accelerates development: Researchers optimized the robot's shape and engine placement simultaneously through iterative simulations and experimental tests, enabling maximum flight control while withstanding extreme thermal and aerodynamic conditions—a process that took two years of dedicated development

• Multi-modal robotics applications emerging: Flying humanoids could revolutionize operations in disaster zones, toxic environments, and missions requiring both aerial mobility and ground manipulation capabilities, with future tests planned at Genoa Airport's dedicated flight area

Source

🤖 Startup Spotlight: Sanctuary AI Need

Location: Paris, France / New York, NY
What They Do: AI development platform expanding into open-source humanoid robotics with affordable, customizable robots and comprehensive development tools
Latest Update: Unveiled two new humanoid robots—HopeJR ($3,000 full-size) and Reachy Mini ($250-300 desktop)—following acquisition of Pollen Robotics, with shipping expected by end of 2025
Why Watch: Hugging Face is applying its successful open-source AI platform strategy to robotics, democratizing access to humanoid development with affordable hardware and comprehensive toolchains, positioning to become the "GitHub of robotics."

3 Key Takeaways:

• Open-source strategy aims to prevent robotics domination by "big players with dangerous black-box systems," leveraging Hugging Face's proven community-driven development model from AI into physical robotics

• Aggressive pricing targets accessibility with HopeJR at $3,000 (vs. competitors at $50K+) and desktop Reachy Mini under $300, potentially democratizing humanoid development for researchers, startups, and hobbyists

• Strategic acquisition of Pollen Robotics provides hardware expertise to complement AI platform capabilities, creating integrated ecosystem from LeRobot development tools to physical robot deployment

Source

📈 Investor Watch

LG CNS Signs Deal with Skild AI to Build Industrial Humanoid Robots

South Korean IT giant LG CNS partnered with US-based Skild AI to develop industrial humanoid robots, marking LG's official entry into the competitive robotics sector. The deal includes both technology collaboration and equity investment through LG Technology Ventures, targeting manufacturing, logistics, and urban services automation.

3 Key Takeaways:

• Mega-round funding momentum: Skild AI is raising a Series B round valuing the company at $4.5 billion, led by SoftBank's $100 million investment, with additional $35 million from NVIDIA ($25M) and Samsung ($10M)—demonstrating massive institutional confidence in robot foundation models

• First Korean corporate entry: LG CNS becomes the first major Korean company to partner with Skild AI, positioning itself against cross-town rivals Samsung Electronics and Hyundai Motor Group who are also heavily investing in humanoid robotics development

• Foundation model approach scales: Skild AI's Robot Foundation Model (RFM) enables robots to interpret multimodal data (images, video, voice, text) for autonomous decision-making, suggesting the same architectural breakthroughs driving LLM success are now being applied to physical AI systems

Source

• 1X Redwood AI Model for NEO Humanoid: End-to-end mobile manipulation system enabling autonomous household task performance through whole-body control, trained on real-world data from 1X's EVE and NEO robots. The model runs fully on NEO's onboard embedded GPU with voice control integration, supporting bi-manual manipulation, adaptive grasping with retry logic, and generalization to unfamiliar objects and task variations.

• Why It Matters: Represents breakthrough in practical home robotics by combining locomotion and manipulation in a single compute-efficient model that operates autonomously on edge hardware. The real-world training approach and ability to handle task variations demonstrates path toward truly general-purpose household robots rather than pre-programmed task execution.

Source

🧩 Pattern of the Week

The Great Scaling Prediction

A new 142-page UBS report featuring 30+ analysts projects explosive growth in humanoid robotics: 2 million robots by 2035, scaling to 300 million by 2050. The total addressable market jumps from $30-50 billion by 2035 to $1.4-1.7 trillion by 2050, spanning components, manufacturing, software, data, and services. This isn't just optimistic forecasting—it reflects a fundamental shift where humanoid robots transition from experimental prototypes to essential labor solutions. The timing aligns with Figure's warehouse breakthroughs, Hexagon's industrial deployments, and massive investment rounds like Skild AI's $4.5 billion valuation. What we're witnessing is the convergence of AI foundation models, simulation-accelerated training, and genuine labor shortages creating the perfect conditions for humanoid robotics to finally achieve commercial viability at scale.

Source

📚 Resource / Reading

When wheels won't do: Humanoid robots for human-centric spaces

This technical deep-dive by NXP Semiconductors' Nicolas Lehment explores why humanoid robots are essential for navigating complex, human-centric environments where traditional wheeled AGVs fail. The piece breaks down three foundational areas—motion control, perception/navigation, and modularity—that are driving adoption beyond structured industrial floors toward truly versatile autonomous systems.

Notable Quotes:

"Modern embodied robots break the old, centralized motor-controller paradigm. Each joint or limb houses a microcontroller responsible for low-latency torque and position loops, while a central processor coordinates full-body motion plans."

"In a simulated kitchen task, a humanoid robot correctly lifted varied vessels 92 percent of the time by combining vision-based detection with force-feedback compensation."

"As these legged platforms become more capable and affordable, we can expect to see them not only in industry, but also in our homes, hospitals, and public spaces, turning the science fiction of years gone by into an everyday reality."

Source

🛠 Builder's Corner

Robots learn welding skills from humans to address welder shortage

University of Nottingham researchers developed a robotic welding system that learns directly from skilled welders by recording their movements and extracting key parameters like torch speed, arc length, and welding angle. The system builds a digital skills library that enables robots to tackle new, complex welding tasks by intelligently combining learned techniques.

What it does: The system uses operation tracking to capture expert welder movements during basic tasks (linear and arc welding), then stores critical parameters (torch traveling speed, welding arc length, welding angle, welding current, wire feeding rate) in a skills library. Robots can then execute complex new tasks like polynomial curves by combining these learned skills.

Why it matters: With half of the UK's welders retiring by 2027 and Brexit limiting EU skilled labor, this approach offers a scalable solution to critical labor shortages. The methodology is expandable to other manual operations like assembly and polishing, potentially transforming how industrial skills are preserved and transferred across multiple sectors while allowing human experts to focus on creative tasks rather than repetitive work.

Source

💼 Jobs in Mechonomics

1. Research Engineer, Robotics – OpenAI

Location: San Francisco, CA (Hybrid) | Salary: $295K-$530K + Equity

Summary: Lead foundational model development for general-purpose robotics in dynamic real-world environments, working across the entire robotics stack from hardware integration to AI models. Role involves owning research agenda for model capability improvements, collaborating cross-functionally, and mentoring team members. Requires research background in robotics, multimodal foundational models, scaling laws, and reinforcement learning, with proven experience managing high-performing research teams.

Source

2. Humanoid Robot Pilot – Figure

Location: Spartanburg, SC (5 days onsite) | Salary: Not disclosed

Summary:Unique hands-on role working directly with learning humanoid robots to teach new behaviors and skills through teleoperation equipment. Position involves wearing teleoperation gear to guide robots through designated tasks, uploading training data to AI systems, and providing daily feedback to the AI team. Requires excellent physical coordination, ability to stand 8+ hours daily, and quick mastery of new physical tasks in a fast-paced industrial environment targeting labor shortages.

Source

3. Robotics Software Engineer – Skild AI

Location: Pittsburgh, PA | Salary: $100K-$300K

Summary: Develop and implement software solutions for general-purpose robotic intelligence systems, focusing on navigation, planning, controls, SLAM, manipulation, and perception. Role involves writing production-level C++ and Python code, collaborating with ML engineers to deploy state-of-the-art models on robots, and working with deployment teams to ensure robust performance across various sites. Requires BS/MS in Computer Science or Robotics with hands-on robot development experience and expertise in ROS/ROS2 platforms.

Source

4. Developer Technologies Engineer, Robotics Reinforcement Learning – NVIDIA

Location: Santa Clara, CA | Salary: $184K-$356.5K + Equity

Summary: Drive adoption of NVIDIA's robot learning tools with developers and robotics manufacturers, working with partner companies on perception-in-the-loop reinforcement learning, learning from demonstration, and multi-agent training. Role involves automating workflows, scaling them in the cloud, collaborating across NVIDIA teams including GEAR for humanoid robots, and running physical applied research experiments. Requires 8+ years software development experience with Python and deep learning stack, plus robotics simulation experience with Isaac Sim, Omniverse, and reinforcement learning.

Source

5. Robotics Engineer, State Estimation, Optimus – Tesla

Location: Palo Alto, CA | Salary: $132K-$390K + Stock Awards

Summary: Design and implement real-time state estimation algorithms for Tesla's humanoid robot Optimus, enabling operation in dynamic, unstructured environments. Role involves developing estimation pipelines that fuse IMU, camera, encoder, and force sensor data for accurate pose and velocity estimation, collaborating with sensor teams on hardware design, and shipping production-quality safety-critical software. Requires modern C++ experience, strong background in state estimation and sensor fusion, plus solid robotics fundamentals in geometry, kinematics, and real-time systems.

Source