Hasty Briefs - Daily Tech News Brief

Bilingual

How Google built its Gemini robotics models
a year ago
- Google DeepMind developed a new family of Gemini Robotics models, specifically designed for robots.
- The models are multimodal, building upon Gemini 2.0 and fine-tuned with robot-specific data to enable physical actions alongside text, video, and audio outputs.
- A bi-arm ALOHA robot successfully performed novel tasks like placing pens inside a shoe and executing a slam dunk with a toy basketball, demonstrating the model's adaptability.
- Gemini Robotics models are highly dextrous, interactive, and general, allowing robots to react to new objects, environments, and instructions without additional training.
- Two main functions are essential for robots: understanding and decision-making (handled by Gemini Robotics-ER) and taking action (handled by Gemini Robotics).
- Gemini Robotics-ER excels in embodied reasoning, detecting objects, and generating code for actions, while Gemini Robotics advances dexterity and multi-step task completion.
- The models adapt to various robot embodiments, from academic robots like ALOHA to humanoid robots like Apollo, enabling diverse applications.
- Potential future applications include complex industrial settings and human-centric spaces like homes, though widespread adoption is still years away.
Remote Access Backdoor Discovered in Chinese Robot Dog Unitree Go1
a year ago
- Security researchers found a pre-installed remote access tunnel in Unitree Go1 robot dogs, allowing full remote control.
- The backdoor was discovered by Andreas Makris and Kevin Finisterre, who reverse-engineered the firmware.
- The tunnel connects to CloudSail, a Chinese remote access platform, enabling unauthorized access to devices.
- Researchers could list connected devices, access web interfaces, use cameras, and log in via SSH with default credentials.
- 1,919 unique Unitree Go1 units were found connected to CloudSail, including devices at MIT, Princeton, and other institutions.
- The tunnel auto-starts on boot and appears deliberately integrated, raising concerns about supply chain trust.
- Recommendations include isolating devices, rotating SSH credentials, and disabling the csclient tunnel service.
Kawasaki unveils a hydrogen-powered, ride-on robot horse
a year ago
- Kawasaki Heavy Industries unveiled the Corleo, a two-seater quadruped robot steered by body movements.
- Powered by a 150cc hydrogen-fueled generator engine, it emits only clean water as a byproduct.
- Features AI vision for navigating rough terrain and a floating seat for rider comfort.
- Designed with grippy rubber treads, adjustable stirrups, and a wind-deflecting screen for high-speed travel.
- AI analyzes terrain and rider's body language, with night-time path visualization via illuminated arrows.
- Currently a CGI concept aimed at 2050 mobility, with a prototype capable of standing and posing.
- Comparable to XPeng's rideable Unicorn, but Kawasaki's design pushes the envelope with high-speed, all-terrain capabilities.
Robotic Gripper Based on Measuring Tape Is Sizing Up Fruit and Veggie Picking
a year ago
- Engineers at UC San Diego developed a robotic gripper called GRIP-tape inspired by measuring tape.
- GRIP-tape is designed for agriculture, capable of gently handling fragile fruits and vegetables.
- The gripper is low-cost, safe around humans, and combines softness with stiffness.
- It features two 'fingers' made of measuring tape spools, controlled by motors for independent movement.
- The gripper can hold various objects, from small fruits to whole vines, and can act as a conveyor belt.
- Future versions may include advanced sensors and AI for autonomous operation.
- The project was partially funded by the National Science Foundation.
A 1980s toy robot arm inspired modern robotics
a year ago
- The Armatron, a 1980s toy robotic arm, inspired modern robotics and AI applications.
- Designed by Hiroyuki Watanabe at Tomy, the Armatron was purely mechanical yet highly functional.
- Watanabe's inspiration came from radio-controlled helicopters, leading to the toy's innovative dual-joystick controls.
- The Armatron was praised by robotics engineers and researchers, bridging the gap between toys and industrial robotics.
- Childhood experiences with the Armatron influenced engineers like Adam Borrell (Boston Dynamics) and Eric Paulos (UC Berkeley).
- Modern robotics and AI still grapple with challenges the Armatron introduced, such as object manipulation and grip optimization.
- The Armatron's legacy continues through online communities, DIY modifications, and its impact on hands-on engineering curiosity.
Micro Wheeled legged Robot
a year ago
- The robot is the smallest two-wheeled-leg robot.
- Includes model files ('OriginalRobotModel.stp') and parts folders for manufacturing and purchasing.
- Four PCBs need fabrication; main control board uses ESP32 with L6234PD013TR motor driver.
- Encoder chip AS5600 and IMU MPU6050 communicate via I2C.
- Servo debugging board unifies serial lines using time-division multiplexing.
- Requires GH1.25 4PIN cables (15cm recommended).
- Arduino IDE-based; brushless motor drive uses simpleFOC.
- WiFi modes: AP (hotspot) and STA (client).
- Power on via XH2.54 plug; initialization includes FOC motor setup.
- Web interface at 192.168.1.11 for remote control (compatible with multiple OS).
Robots race in Chinese half marathon
a year ago
- Robots participated in the Yizhuang half marathon in Beijing alongside human runners.
- Twenty-one humanoid robots designed by Chinese manufacturers took part in the 21km race.
- Tiangong Ultra won the race with a time of 2 hours, 40 minutes, and 42 seconds.
- Some robots completed the race, while others faced difficulties, including one that fell at the start.
- The event was promoted as the world's first robot half marathon.
Chinese robots lose against humans in half-marathon
a year ago
- Robot Tiangong Ultra completed a half-marathon in 2 hours 40 minutes but was slower than the fastest human time of 1 hour, 11 minutes, 7 seconds.
- The event featured 21 bipedal robots racing alongside 10,000 humans, showcasing various designs from Chinese manufacturers.
- Robots received batteries and technical tools at aid stations, while humans had water and snacks.
- The race was a technical demonstration, with no robot expected to win.
- China is investing in AI and robotics to boost technological strength and economic growth, though some question the practical utility of such demonstrations.
Most Humanoid Robots Fail to Finish Half Marathon in Beijing
a year ago
- 12,000 human athletes and 21 humanoid robots participated in a half-marathon in Beijing, marking the first time humans and bipedal robots ran in the same race.
- Only six robots finished the race, with the fastest, Tiangong Ultra, completing it in 2 hours and 40 minutes after multiple battery changes and a fall.
- Robotics expert Alan Fern notes that humanoid robots are designed for task completion in diverse environments, not speed, and highlights hardware robustness as a key achievement.
- The race exposed limitations in China's humanoid robot industry, with most robots overheating, falling, or requiring frequent replacements.
- Robots like Tiangong Ultra and N2 stood out for their consistent pace, while others, like Huanhuan and Shennong, performed poorly or malfunctioned.
- Human operators played a crucial role, guiding robots, changing batteries, and using duct tape for repairs during the race.
- The event showcased design flaws more than capabilities, but still highlighted progress in bipedal robot hardware.
Tiny, hackable telepresence robot for under $100
a year ago
- Goby is a small, reprogrammable robotic camera platform by Charmed Labs, crowdfunding for under $100.
- Features 'tinypresence' for a unique first-person perspective in small-scale environments like LEGO arenas or tabletop games.
- Includes an articulated caster tail for pan/tilt without a gimbal and easy recharging via a disc-shaped charger.
- Powered by ESP32-S3 and OmniVision OV2640, streaming live video via WebRTC, with encrypted connection setup via URL/QR code.
- BitBang code (in beta) simplifies remote access for ESP32 devices, planned for open release.
- Potential creative uses discussed, including as a miniature FPV device, though not officially endorsed.
- Criticism includes outdated MicroUSB port, but praised for hackability and affordability.
π0.5: A VLA with open-world generalization
a year ago
- Robots have advanced significantly, performing complex tasks like folding laundry and cleaning tables.
- The biggest challenge in robotics is generalization—adapting to new settings and objects.
- Generalization requires robust physical skills and common-sense understanding of the environment.
- Most commercial robots operate in controlled environments like factories due to limited generalization.
- Robotic foundation models like π0.5 aim to generalize to messy, real-world environments.
- π0.5 can perform tasks in new homes, showing flexibility and resourcefulness.
- Co-training on heterogeneous data enables π0.5 to understand semantic context and transfer skills.
- π0.5 combines high-level semantic decisions with low-level motor control for complex tasks.
- Experiments show π0.5 can clean kitchens and bedrooms in unseen environments.
- Future improvements include autonomous learning and better knowledge transfer.
More than 20 robots race alongside humans at half marathon in Beijing
a year ago
- More than 20 robots participated in a half marathon in Beijing alongside human runners.
- Robots varied in size, with some as short as 120 cm and others as tall as 1.8 meters.
- Tiangong Ultra, the winning robot, completed the race in 2 hours and 40 minutes, aided by long legs and a specialized algorithm.
- Some robots struggled during the race, with incidents like falling at the start or crashing into railings.
- This event marked the first time humanoid robots raced alongside humans in a 21-kilometer marathon.
- China views robotics as a frontier industry that could drive future economic growth.
- Experts question whether marathon participation is a reliable indicator of robots' industrial potential.
A Review of the Personal Humanoid Robots
a year ago
- Humanoid robots are set to enter homes by 2025, transforming daily life with tasks like chores, tutoring, and elderly care.
- The world is designed for humans, making humanoid robots the most efficient form for automation in human-centric spaces.
- Humanoid robots offer advantages like using existing tools, navigating human environments, and fostering trust through anthropomorphic design.
- Economic benefits include scalability and reprogrammability for multiple roles, reducing the need for specialized robots.
- Psychological factors favor humanoid robots for acceptance in customer-facing and home settings.
- Challenges include complexity, cost, and the 'uncanny valley' effect, but these are expected to diminish with technological advancements.
- A historical timeline shows the evolution from industrial robots to personal robots like the Roomba and Tesla's Optimus.
- The development of humanoid robots is likened to a 'Sputnik moment,' with global competition among nations and companies.
- Security concerns arise as robots connected to the internet could be hijacked for malicious purposes.
- China is aggressively investing in robotics, aiming to dominate the market by 2027, while the U.S. focuses on innovation.
- The global value of humanoid robots is estimated at up to $24 trillion, with significant potential in household and manufacturing applications.
- Technological advancements are reducing costs, making robots more accessible, with prices expected to drop below $10,000 by 2026.
- Leading companies like Tesla, Figure AI, and 1X Technologies are developing robots for various applications, from kitchen assistance to companionship.
- Adoption of robots could address labor shortages, reduce costs of goods and services, and increase leisure time.
- Emerging trends include organized aggression against robots and the rise of a 'Robot Tax' in the European Union.
- The article offers exclusive insights for members on the cultural, philosophical, and practical impacts of humanoid robots.
Hugging Face to sell open-source robots thanks to Pollen Robotics acquisition
a year ago
- Hugging Face acquires Pollen Robotics to expand into open-source robotics.
- Reachy 2, an open-source humanoid robot, is now available for $70,000.
- LeRobot library, launched in 2024, has grown to over 12,000 GitHub stars.
- Pollen Robotics' Reachy 2 features advanced hardware, VR compatibility, and open-source software.
- Hugging Face's fifth acquisition includes notable additions like Gradio and XetHub.
- NVIDIA collaborates with Hugging Face to accelerate LeRobot's AI workflows.
- Reachy 2 is used in institutions like Cornell and Carnegie Mellon for AI research.
- Pollen Robotics has deployed hundreds of robots across 20+ countries.
- Hugging Face aims to democratize robotics with affordable, open solutions.
Robot Dexterity Still Seems Hard
a year ago
- Humanoid robots are being developed by numerous companies, with significant venture capital investment.
- Existing companies like Tesla, Boston Dynamics, and Chinese EV manufacturers are also entering the humanoid robot space.
- Humanoid robots are becoming more capable, with demonstrations of running, dancing, and performing complex tasks.
- Despite impressive locomotion, manipulation remains a significant challenge for humanoid robots.
- Robots struggle with dexterity, which involves manipulating a variety of objects in different ways quickly and flexibly.
- Moravec's Paradox highlights that tasks easy for humans, like unwrapping a bandaid, are difficult for robots.
- Current humanoid robots show limited dexterity, often struggling with precise and subtle movements.
- Robotic hands lack the strength, sensitivity, and tactile feedback of human hands, hindering dexterity.
- Dexterity issues are both hardware and software problems, with current robotic manipulators being expensive and limited.
- Proprioception, or the sense of body position and movement, is another critical challenge for robotic dexterity.
- Tactile sensor density and efficacy are major hurdles, as robots lack the multi-channel, analog feedback of human touch.
- Specialized robots may overcome some dexterity limitations, but general-purpose humanoid robots face significant barriers.
- Progress in robotic dexterity may follow a slow, incremental path similar to self-driving cars.
- Humanoid robots may not be the optimal form factor for all tasks, with other designs potentially being more effective.
Show HN: Kinematic Hand Skeleton Optimization in Jax
a year ago
- A repository for training robots using Pi0 and Lerobot with human pose motion retargeting.
- Currently supports only Linux and requires the Pixi package manager.
- Commands to clone the repo and navigate into it.
- Instructions for obtaining camera intrinsics/extrinsics, depth maps, and point clouds from multiview images.
- Commands for camera calibration and dataset visualization for HOCap and Assembly101 datasets.
- Pipeline details for 2D pose estimation, tracking, and triangulation.
- Goal: Convert multi-camera footage into axis-angle joint angles and 3D positions for robotic retargeting.
- Benefits of using a kinematic skeleton with joint angles.
- Steps in the Capture-to-Angles Pipeline.
- Commands to run notebook tutorials for various functionalities.
- Future plans include adding egocentric views, pose retargeting to Isaac Sim, and fine-tuning on simulators.
Achieving Human Level Competitive Robot Table Tennis
a year ago
- Research on achieving human-level competitive robot table tennis.
- Collaboration by multiple authors from Google DeepMind and other contributors.
- Primary and core contributors are listed with their affiliations.
- Equal contribution noted among corresponding authors.
RUKA: Rethinking the Design of Humanoid Hands with Learning
a year ago
- Dexterous manipulation is a fundamental capability for robotic systems, but progress has been limited by hardware trade-offs.
- RUKA is a tendon-driven humanoid hand that is compact, affordable, and capable, made from 3D-printed parts and off-the-shelf components.
- RUKA features 5 fingers with 15 underactuated degrees of freedom, enabling diverse human-like grasps.
- The hand's tendon-driven actuation allows powerful grasping in a compact, human-sized form factor.
- Control challenges are addressed by learning joint-to-actuator and fingertip-to-actuator models from motion-capture data collected by the MANUS glove.
- Extensive evaluations show RUKA's superior reachability, durability, and strength compared to other robotic hands.
- Tele-operation tasks demonstrate RUKA's dexterous movements.
- The open-source design, assembly instructions, code, and data for RUKA are available on the project's website.
Amazon makes 'fundamental leap forward in robotics' device having sense of touch
a year ago
- Amazon has developed a new robot named Vulcan with a sense of touch, capable of handling about 75% of items in its warehouses.
- Vulcan uses AI to identify objects by touch, determining what it can handle and the best way to pick them up.
- The robot will work alongside humans, reducing the need for ladders or frequent bending by stowing items on upper and lower shelving levels.
- Amazon's director of robotics, Aaron Parness, describes Vulcan as a 'fundamental leap forward in robotics' due to its tactile capabilities.
- The deployment of Vulcan raises concerns about job losses as automation increases in distribution centers.
- Goldman Sachs economists speculated that 300 million jobs worldwide could be automated by 2030 due to AI advancements.
- Amazon's chief technologist of robotics, Tye Brady, asserts that robots cannot fully replace humans and are meant to 'amplify human potential' and improve safety.
- Brady emphasizes that humans are essential for detecting system hacks and spotting practical issues like broken or spilled items.
- AI is helping robots navigate complex spaces autonomously and learn to move safely alongside people and objects.
- Amazon plans to install machine-learning-based packaging technology in Europe to reduce waste, with over 70 machines set up by the end of 2024.
System lets robots identify an object's properties through handling
a year ago
- Researchers from MIT, Amazon Robotics, and the University of British Columbia developed a technique enabling robots to estimate object properties (e.g., mass, softness) by shaking them, using only internal sensors.
- The method eliminates the need for external tools like cameras, making it cost-effective and useful in low-visibility environments (e.g., dark basements, post-earthquake rubble).
- Key innovation: A simulation process combining robot and object models to rapidly identify characteristics during interaction, leveraging proprioception (joint encoder data).
- The approach matches or outperforms vision-based methods in accuracy and is robust to unseen scenarios, requiring minimal real-world data.
- Future goals include combining with computer vision for multimodal sensing, expanding to complex systems (soft robots, granular materials), and enhancing adaptive robot learning.

Hasty Briefsbeta

#robotics

How Google built its Gemini robotics models

Remote Access Backdoor Discovered in Chinese Robot Dog Unitree Go1

Kawasaki unveils a hydrogen-powered, ride-on robot horse

Robotic Gripper Based on Measuring Tape Is Sizing Up Fruit and Veggie Picking

A 1980s toy robot arm inspired modern robotics

Micro Wheeled legged Robot

Robots race in Chinese half marathon

Chinese robots lose against humans in half-marathon

Most Humanoid Robots Fail to Finish Half Marathon in Beijing

Tiny, hackable telepresence robot for under $100

π0.5: A VLA with open-world generalization

More than 20 robots race alongside humans at half marathon in Beijing

A Review of the Personal Humanoid Robots

Hugging Face to sell open-source robots thanks to Pollen Robotics acquisition

Robot Dexterity Still Seems Hard

Show HN: Kinematic Hand Skeleton Optimization in Jax

Achieving Human Level Competitive Robot Table Tennis

RUKA: Rethinking the Design of Humanoid Hands with Learning

Amazon makes 'fundamental leap forward in robotics' device having sense of touch

System lets robots identify an object's properties through handling