- The Rundown Robotics
- Posts
- Berkeley's open-source $5K humanoid
Good morning, robotics enthusiasts. UC Berkeley just unveiled a fully open-source humanoid that can be built with 3D-printed parts for as little as $5K.
It walks, writes, and learns tasks through AI-driven reinforcement learning. With projects like these gaining momentum, the open-source humanoid movement is picking up speed. The question is: Are you ready to build your own DIY humanoid?
In today’s robotics rundown:
UC Berkeley’s open-source humanoid
Hugging Face’s 3D-printed robot arm
Researchers make a robot cake you can eat
Sewer inspection drones do the dirty work
Quick hits on other robotics news
LATEST DEVELOPMENTS
UC BERKELEY
Image source: University of California, Berkeley
The Rundown: Researchers at UC Berkeley just introduced Berkeley Humanoid Lite, a lightweight humanoid version of its research-geared open-source bot that can be built at home with 3D printed parts for less than $5K.
The details:
Berkeley Humanoid Lite can walk, solve a Rubik’s Cube, write with a marker, and perform real-world tasks using AI-driven reinforcement learning.
It weighs 16 kg, stands nearly 3 feet tall, and features modular, 3D-printed cycloidal gearboxes for its actuators and frame.
An onboard 6S 4000 mAh LiPo battery powers the robot for about 30 minutes, with optional tethering also available for extended sessions.
The whole design leverages widely available electronic and mechanical parts, making it easy for researchers and hobbyists to build, modify, and repair.
Why it matters: All hardware designs, code, and frameworks for the humanoid are fully open-source, and the researchers have even opened their WhatsApp communities for collaboration and support, making DIY robotics more accessible than ever. They also report that in testing, the bot proved as durable as robots costing up to $100K.
HUGGING FACE
Image source: Hugging Face
The Rundown: Open-source leader Hugging Face is ready to give you a hand—literally. The AI startup just introduced the SO-101, a 3D-printable, programmable robotic arm that performs basic tasks through reinforcement learning.
The details:
The SO-101 arm features six motors and can be equipped with cameras, enabling it to perform tasks like picking up and sorting objects.
Building on its predecessor, the SO-100, it uses a leader-follower system for intuitive teleoperation when handling different objects.
All instructions, 3D files, and part lists are openly available on GitHub, allowing anyone to build their own robotic arm at a starting cost of around $100.
The SO-101 is also compatible with NVIDIA Jetson AI and can be enhanced with mobile bases for greater autonomy, supporting robotics research.
Why it matters: This is a major step up from last year’s S0-100, which was Hugging Face’s first venture into open-source robotics. Since then, the startup has come a long way — acquiring Pollen Robotics and expanding its portfolio as it works to democratize robotics through a collaborative, open-source approach.
EPFL
Image source: EPFL
The Rundown: Researchers from the École Polytechnique Fédérale de Lausanne (EPFL) and the Italian Institute of Technology created RoboCake—a fully edible, robotic wedding cake that also comes with edible batteries.
The details:
The cake, showcased at Expo 2025 Osaka, features two edible, pomegranate-flavored gummy bears that move via an internal pneumatic system.
The world’s first edible, rechargeable battery—made of vitamin B2, quercetin, activated carbon, and chocolate—powers LED candles on the cake.
The cake is part of the EU-funded RoboFood project, a four-year, 3.5M euro research project to create edible robots and robotic foods.
Another major edible robotic project is ELFO, which is developing edible electronic circuits and sensors for applications like food supply chain tracing.
Why it matters: Novelty aside, edible robots can have life-saving applications such as smart pills, ingestible sensors for noninvasive diagnosis, and wildlife preservation (imagine robots delivering medicines to dangerous animals). Plus, it’s a big opportunity with the global edible soft robotics market projected to reach $49.2B by 2029.
FLYBOTIX
Image source: Flybotix
The Rundown: Back in 2016, a massive sinkhole opened in Fraser, Michigan, triggering the collapse of a major underground sewer line. In the aftermath, the region has become a leader in using drones and AI for inspecting aging sewer systems.
The details:
Wired reports that Macomb County, Michigan, is using Flybotix’s Asio X drone and SewerAI software, which can inspect up to 900 meters of pipe per day.
The drone features a protective cage, high-powered LED lights, and a 4K camera to capture footage inside pipes buried up to 60 feet underground.
The AI quickly analyzes the footage from the drone, automatically identifying defects such as cracks or blockages — and prioritizing repairs.
The total investment for the drone and AI system was about $100K, compared to the $1M previously spent every three years on manual inspection.
Why it matters: Inspecting old sewer systems is dangerous, dirty work, and drones make the process safer, faster, and more cost-effective. Atlanta and Detroit use drones equipped with LiDAR and 3D imaging to map aging sewer systems, while Houston has replaced 80% of manned inspections with drones, saving workers from gas exposure.
QUICK HITS
📰 Everything else in robotics today
China’s UBTech Robotics reportedly secured a buyer for a “small batch” of its humanoids to be deployed in manufacturing.
An alleged video of Unitree’s H1 humanoid exhibiting erratic behavior and nearly injuring two workers has gone viral.
Unitree also unveiled fire rescue robot dogs that can shoot water or foam with a range of 60 meters and a flow of 40 liters per second.
Near Space Labs nabbed $20M in Series B funding to expand deployment of its Swift robots for capturing high-res images in the stratosphere.
Montreal opened doors to Canada’s first and only hospital-based center dedicated to advancing surgical robots.
University of Houston researchers developed MyoStep, a next-gen soft exoskeleton designed to help children with cerebral palsy walk more easily and confidently.
Northeastern University researchers developed a novel algorithm, DFLIOM, that boosts mobile robot navigation by relying only on essential data points for 3D mapping.
Swiss researchers are developing LEVA, a multi-terrain autonomous delivery bot that can lift and drop off cargo boxes weighing up to 85 kg.
Trump’s tariffs were reportedly “top of mind” at last week’s Robotics Summits in Boston, with thousands of tech industry professionals attending.
The Netherlands-based Lely dairy farm robot milks cows when the cows choose to visit the bot for milking, while also handling feed and manure cleanup.
The UK’s National Health Service approved 11 cutting-edge robotic surgery systems for soft tissue and orthopedic procedures.
China unveiled a tube-shaped, two-pound grenade drone that allows a single soldier to control multiple units via AI targeting assistance.
British YouTuber James Bruton created a DIY unicycle robot that can balance and move forward and backward on one large wheel.
COMMUNITY
Join our next workshop on Tuesday, May 6th at 3 PM EST with Chris Cunningham and Joe Zapiain from ClickUp. In this live session, you’ll learn how to embed AI and automation directly into your ClickUp workspace to save time, reduce complexity, and scale your team's execution, without adding headcount.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
That's it for today!Before you go we'd love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
See you soon,
Rowan, Jennifer, and Joey—The Rundown’s editorial team