Bilingual

Segmenting Robot Video into Actionable Subtasks
25 days ago
- Introducing WGO-Bench, a benchmark for robotics subtask annotation across 100 video episodes with 743 segments covering 62 task instructions.
- Over 60 experiments identified the best subtask annotation pipeline: best segmentation F1 of 0.306, labeling accuracy of 61.0%, and end-to-end F1 of 0.168.
- Gemini models, particularly Gemini 3.5 Flash, outperformed other models by 24.5% and are best for this task.
- Cost-effective method uses contact sheets, costing $2.64 per hour of video (batch pricing), roughly 19x cheaper than human annotation.
- Key techniques include visual timestamping on contact sheets, strict annotation protocols, and using previous/current/next segment context for labeling.
- Segmentation is the main bottleneck, especially for short subtasks, while labeling improves with contextual visual input.
- Open-sourced pipeline in Refiner for public use, enabling scalable subtask annotation without human intervention.
AI is 'not smart' so what's next in artificial intelligence?
22 days ago
- Yann LeCun critiques current LLMs (like ChatGPT) for lacking real-world understanding and physical reasoning, comparing them unfavorably to even animals like rats.
- He left Meta to found AMI Labs, aiming to develop AI beyond LLMs, focusing on flexible systems for complex real-world tasks such as robotics and household chores.
- AMI Labs raised over $1bn in seed funding, with backers including Nvidia and Jeff Bezos's fund, highlighting investor confidence in next-generation AI.
- LeCun's approach, Joint Embedding Predictive Architecture (JEPA), creates abstractions to filter useless information and predict outcomes without over-predicting, like not guessing a pen's fall direction.
- Other researchers, like Ingmar Posner at Oxford, are working on 'World Models' that simulate environments to enable causal reasoning and explainable AI, inspired by advances since 2018.
- Projects like Google's Dreamer, DeepMind's Genie, and Wayve's Gaia exemplify the shift towards AI that learns from mental simulations, with applications in gaming and robotics.
- LeCun predicts AMI Labs' model will be tested in industrial settings soon, with long-term goals for general intelligence requiring minimal training.
- Despite AI advancements, humans will remain crucial for asking questions, creativity, and leadership, with AI acting as smarter assistants.
Heuristics for lab robotics, and where its future may go
21 days ago
- There are two main types of lab robots: 'boxes' (like liquid handlers) that automate specific tasks and 'arms' (like the UR5) that move materials between boxes to form integrated workcells.
- Most lab protocols can be automated, but often aren't worth automating because the upfront investment in time and cost is only justified for high-throughput, repetitive tasks, not for exploratory research that changes frequently.
- Three approaches to improve lab automation are: better translation layers (e.g., Synthace, Briefly Bio) to convert human intent into robot instructions, better physical hardware integration (e.g., Automata, Ginkgo Bioworks) to connect systems, and adding intelligence (e.g., Medra, Zeon Systems) for error recovery and adaptability.
- The long-term business model for lab automation likely converges on 'cloud labs' (like the defunct Transcriptic or Emerald Cloud Labs), which centralize experiments to maximize robot utilization and can vertically integrate to reduce costs, though early attempts struggled by being too general and offloading logistics to users.
- The ultimate value of lab robotics may not be just increasing throughput, but enabling better, more predictive assays (improving 'predictive validity') and higher-standard experimentation at scale, potentially transforming drug discovery.
Drone Autonomy Crash Course
20 days ago
- A short guide to drone autonomy focusing on modeling, state estimation, motion planning, and control for quadcopters.
- Aimed at newcomers to provide an intuitive overview and serve as a 'jumping off point' with references for deeper study.
- Based on the author's experience with the challenge of understanding academic papers on quadcopter autonomy.
- Originally a long paper but broken into digestible pieces with casual writing style for accessibility.
- Includes incomplete sections (WIP) to be updated soon, emphasizing practical understanding over rigor.
China wants to solve the hardest problem in robotics – making hands
18 days ago
- Human hands are highly complex and involve intricate neurological control for simple tasks.
- AI and robotics are progressing in China to replicate human-like dexterity for practical uses.
- Chinese startups are leveraging manufacturing and 'embodied AI' to develop advanced robotic hands.
- Robotics is seen as key to China's economic future amid workforce challenges.
- China leads in factory robot deployment, but multipurpose humanoid robots are still limited.
- Robotic hands are a major engineering challenge, requiring sophisticated hardware and software.
- China's supply chain and EV industry expertise are advancing hardware components for robots.
- Data collection for training robots involves teleoperation and sensor-based methods like smart gloves.
- Entrepreneurs aim to make robots affordable and useful, such as through low-cost prosthetic hands.
- The goal is to enhance human life with robots, not replace labor, fostering prosperity.
Robots, Soft Power, and Summer Davos 2026
17 days ago
- China is the world's dominant industrial robotics market, accounting for 54% of new installations in 2024 with 295,000 units.
- The 2026 Summer Davos in Dalian included direct discussions on labor issues like wage gaps and trade policy, alongside 'soft' robot art exhibits.
- Media coverage naturally favored shareable robot demonstrations over policy sessions, reflecting attention economics rather than deliberate suppression.
- China's automation drive is structurally motivated by rising labor costs and an aging population, framed as innovation leadership.
- Comparatively, major economies like the U.S. also emphasize optimistic automation narratives, indicating a global pattern.
- Analysis found no evidence that cultural programming was used to avoid scrutiny of labor conditions at the event.
- Unverified hypotheses include potential curatorial bias and undisclosed sponsorship, requiring further targeted research.
Mistral's Robostral Navigate: a state of the art robotics navigation model
16 days ago
- Robostral Navigate is an 8B model for autonomous robot navigation using only a single RGB camera.
- It achieves 76.6% success on unseen R2R-CE benchmarks, outperforming multi-sensor approaches.
- The model combines pointing-based navigation with reinforcement learning for continuous improvement.
- Built entirely in-house with simulation-trained data and token-efficient training techniques.
- It generalizes across robot types (wheeled, legged, flying) and adapts to real-world obstacles.
- Efficient training via prefix-caching reduces token use by 22x, speeding up training significantly.
- Online reinforcement learning (CISPO) boosts performance by 3.2% and enables recovery from failures.
- The technology unlocks applications in manufacturing, delivery, logistics, and hospitality.
- Robostral Navigate is a step toward unified embodied AI, paving the way for general-purpose robotics.
NEO's Hands – An API to the Physical World
15 days ago
- 1X announces Bill Nash as new CFO, bringing financial leadership experience from ambitious tech companies.
- Breakthrough 25 Degree of Freedom (DOF) tendon-driven hands for NEO humanoid platform achieve near human-level dexterity, strength, safety, and reliability.
- These hands are designed to remove hardware limitations, making data the only barrier to robot capabilities.
- NEO can perform tasks like assembling LEGOs, using tools, grasping delicate objects, and communicating via sign language.
- The hands feature force transparency, backdrivability, and high-resolution tactile sensing for real-time feedback.
- They are IP68 waterproof, food-safe, and built for durability with millions of test cycles under high loads.
- Deep vertical integration of motors, sensors, and tendon systems enables rapid iteration and performance.
- Production capacity aims for 10,000 hands this year, scalable to support large-scale data collection for AGI development.
Show HN: Leet Robotics: Learn robotics and ROS2 with hands-on courses
9 days ago
- The platform offers a free, accessible robotics learning environment with no account required to start.
- Users can pick a course, use an in-browser workspace with ROS2, Gazebo, and Foxglove, and begin coding and simulating robots.
- Featured courses include 'Intro to ROS2' where learners build an autonomous museum-guide robot, covering fundamentals like nodes, topics, TF, and obstacle avoidance.
- Interactive lessons with drag-and-play elements are designed to enhance understanding through hands-on experimentation.
- The cloud workspace allows building, simulating, and visualizing robotics projects entirely in the browser, with no setup or hardware needed.
A General Goal-Conditioned Minecraft Model
9 days ago
- Pantograph develops a method for learning goal-directed behavior through pretraining on internet-scale video, improving ability to achieve complex goals.
- The Pan model (4B parameters) in Minecraft shows strong performance in fighting, exploration, building, and handling diverse environments.
- Goal-conditioned reinforcement learning uses future video frames as goals, side-stepping the need for explicit reward functions.
- Pretraining uses 500k hours of Minecraft video; post-training uses 2k hours of action-labeled contractor data for policy learning.
- Evaluation on 104 environments shows Pan models outperform STEVE-1 and VLA baselines in goal-directed tasks and generalization.
- Scale matters: Pan-4B significantly outperforms smaller versions in building and mechanisms, with notable successes and failures.
- Models exhibit interesting reward hacking behaviors, finding unintended ways to achieve goal images in offline training.
- Future work includes scaling models, broadening data distribution, and using models for online RL and real-world robotics.
1X's NEO hand is genuinely impressive. The economics are brutal
10 days ago
- 1X's NEO hand features 25 degrees of freedom, tendon-driven actuation, and force transparency, enabling advanced manipulation like assembling LEGO, playing video games, and zipping jackets.
- The hand's design avoids 'write-only' issues through low gear ratios (5:1 to 15:1), allowing it to sense touch and force, with tactile sensors that detect slip and shear for precise control.
- 1X uses teleoperation for some demos, with plans for Scheduled Expert Mode, but this model could incur high labor costs, estimated at over $100 million annually for 10,000 units.
- Priced at $20,000 or $499/month, NEO likely sells at a loss per unit, with potential hardware losses in hundreds of millions, despite raising $137 million, far less than competitors like Figure AI ($1 billion+).
- 1X bets on vertical integration, manufacturing components in-house to reduce costs and improve capabilities, but faces challenges in reliability verification, software development, and funding constraints.
- The hand is a breakthrough in robotics, setting a new standard for humanoid robots, but 1X's business viability depends on scaling production, managing costs, and advancing AI to match the hardware.
Camera Chase Vehicle
15 days ago
- The project involves building a cinematic chase camera rover using a large 1/5 scale RC chassis found at an industrial auction.
- Key components include a Movi M10 3-axis gimbal for stabilization, a VESC motor controller, and a Connex Amimon wireless HD video system for low-latency transmission.
- Custom parts like TPU fenders, a structural gimbal mount, and battery mounts were 3D printed or CNC machined to fit the chassis and improve durability.
- Power is supplied by two DeWalt 20V battery packs in series for the chassis and a separate ~14V DTAP battery for the gimbal and camera.
- Initial testing revealed gimbal balancing issues and mechanical failures, leading to upgrades like stiffer springs and flexible TPU parts to absorb impacts.
- The rover was successfully tested in ice racing conditions, capturing stabilized footage of vehicles like karts, and survived abuse including a tumble at speed.
- Camera upgrades from a Panasonic GH4 to a GH5s improved image quality, and remote lens control was implemented using a digi-pot and microcontroller.
- CAD files and 3D print designs for custom parts are shared for replication.
New spinning drone hides in plain sight
7 days ago
- Northwestern University engineers designed a drone called 'Phantom Twist' that uses motion blur to appear nearly invisible.
- The drone spins at 25 times per second, too fast for the human eye to see clearly, creating a ghostly smudge that blends into the background.
- This approach focuses on exploiting human perception of motion rather than using camouflage or transparent materials.
- Potential applications include wildlife monitoring, environmental surveys, and infrastructure inspection with reduced visual disruption.
- The research was presented at Robotics: Science and Systems 2026 in Sydney, Australia, led by Professor Michael Rubenstein.

Hasty Briefsbeta

#robotics

Segmenting Robot Video into Actionable Subtasks

AI is 'not smart' so what's next in artificial intelligence?

Heuristics for lab robotics, and where its future may go

Drone Autonomy Crash Course

China wants to solve the hardest problem in robotics – making hands

Robots, Soft Power, and Summer Davos 2026

Mistral's Robostral Navigate: a state of the art robotics navigation model

NEO's Hands – An API to the Physical World

Show HN: Leet Robotics: Learn robotics and ROS2 with hands-on courses

A General Goal-Conditioned Minecraft Model

1X's NEO hand is genuinely impressive. The economics are brutal

Camera Chase Vehicle

New spinning drone hides in plain sight