🤖 One Prompt Gives Pumpkins Legs
Good Morning, Roboticists!
We spent thousands of years carving faces into pumpkins. Now they're growing legs. Halloween has officially become a two-way threat.
One Prompt Gives Pumpkins Legs
TL;DR: Google DeepMind showed a physical dancing pumpkin robot with twelve degrees of freedom, saying AI agents designed and simulated it from one plain-English prompt without manual CAD. Google identified Gemini 4 Argon and Antigravity as the tools behind the project. The Halloween demonstration makes agent-assisted mechanical design tangible, without establishing autonomous fabrication, production readiness or general robot design reliability. Read more →
Robot Fixes Lose 13.6 Days
TL;DR: HD Hyundai Robotics says AI agents developed with Palantir cut average robot problem-solving time from 21.5 days to 7.9 days after their August introduction. Agents inspect existing programs and suggest changes that engineers refine. The newly disclosed results put numbers behind factory programming assistance, while a projected reduction of more than half in programming costs remains an expectation. Read more →
Meta Borrows A Robot Body
TL;DR: Meta has built a humanoid robot prototype with Chinese startup Fourier and is developing a second version, The Information reported, according to Reuters. The report adds a concrete hardware partner to Meta’s robotics ambitions, but Reuters could not independently verify it. No task results, deployment figures or commercial timetable are established by the brief wire report. Read more →
PUBG Wants Robots Reading People
TL;DR: KRAFTON’s Ludo Robotics presented Ludi 0.2 at UC Berkeley, pushing its robot intelligence toward reading gestures, remembering context and responding when people change their instructions. The update links conversation, navigation and manipulation within supported environments. It is a research demonstration rather than a consumer launch, and the announcement provides no success rates, response times or delivery plans. Read more →
Lumi Clears A Conditional Gate
TL;DR: Lightberry’s Lumi received a conditional exemption from the FCC Covered List, opening a route to the American market for its interactive humanoid. The company says it replaces onboard sensing, computing and software and is moving manufacturing to the United States. The exemption depends on continued compliance and vetting; it does not establish deployment scale or unrestricted government certification. Read more →
The Cook Changes Its Tools
TL;DR: SwapBot gives robot arms interchangeable kitchen tools with tactile sensing, letting them swap attachments in about ten seconds. University of Washington researchers trained policies for cutting, brushing and picking tofu. Across twenty complete cooking trials, tactile policies succeeded 60% of the time, versus 35% using cameras alone. Slippage and unfamiliar states still interrupted the meal. Read more →
This Drone Grabs While Flying
TL;DR: CAPTURER combines a flying platform, robotic arm and onboard vision to grasp objects without external localization. Researchers tested stationary and moving targets indoors and outdoors, including picking a marker pen just 1.5 centimeters wide at flight speeds above 1.5 meters per second. Coordinated planning and disturbance compensation make the aircraft an active manipulator, with precision demonstrated in controlled experiments. Read more →
One Magnet Commands Sixteen Shapes
TL;DR: Princeton researchers can steer a four-cell origami robot through sixteen stable configurations using one attached magnet and a programmed external magnetic field. Different folding thresholds let pulses select individual cells, while a reset sequence restores a known starting shape. The result brings mechanical memory and control together; proposed medical transport would still require substantial miniaturization. Read more →
Robot Hands Borrow Human Touch
TL;DR: TouchBridge records human hand movements and fingertip contact, then converts those demonstrations into training data for robot hands. Across four physical tasks, policies trained on converted examples averaged 72.5% success, compared with 77.5% using equal numbers of teleoperated robot examples. Human demonstrations were collected seven times faster, suggesting a cheaper data pipeline while leaving a measurable execution gap. Read more →
Robot Muscles Snap Into Bricks
TL;DR: MUSBRIC turns pneumatic artificial muscles into reusable bricks that connect to standard LEGO components. Researchers paired the hardware with a calibrated simulator, tested fifteen leg configurations and assembled a hand and walking quadruped. The platform makes changing muscle attachments a construction task rather than repeated fabrication. Hardware designs and simulation code are promised for release, rather than already available. Read more →
TOGETHER WITH US
AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.
We've helped promote over 500 Tech Brands. Will yours be the next?
Email our co-founder Mark directly at mark@aisecret.us if the button fails.