Reachy Mini Desktop Robot Gets All-local, Conversational AI

Reachy Mini is a limbless desktop robot from Hugging Face made for human interaction experiments, and to give you an idea of what it’s like is a guide on how to implement expressive, local conversational AI complete with head movements and antenna wiggles. It’s conversational in the sense that it aims to feel natural, with low-latency responses and the ability to interrupt, with everything running on local hardware if one so wishes.

Reachy Mini can use remote services, or work in tandem with a desktop machine or laptop.

The software stack is essentially VAD (voice activity detection) → STT (speech-to-text) → LLM (large language model) → TTS (text-to-speech) which allows users to tweak things to their liking, or independently swap or modify pieces as things evolve.

This also allows users to tailor the services to match whatever their hardware is capable of. For example, one could easily use a frontier AI model via remote API for the LLM while keeping everything else local.

The local models in the example configuration are effective and relatively modest (Qwen3-4B-Instruct for the LLM, and even smaller models for the rest) but it’s nice to have the option to offload parts to remote providers if necessary.

Reachy Mini looked very interesting when it was launched as a kit last year, and since then Hugging Face has built up an impressive software suite and infrastructure through which users can easily share their applications. If you’re curious, there’s a simulator for Reachy Mini which should give you an idea of what it can do.

3D Printed Robot Arm Built For Learning Purposes

If you want to work with robots you can do all sorts of learning with software and simulation, but nothing quite beats getting to grips with real machinery. That was the motivation for [James Gullberg] to build this impressive robot arm. 

Featuring six degrees of freedom, the robot arm is mostly constructed of 3D printed components. This let [James] experiment with a wide variety of joint and reducer designs for the sake of learning and investigation. The base of the robot uses a fairly conventional planetary gear drive, while shoulder and elbow joints rely on split-ring planetary gearboxes to allow for high torque density with regards to size. [James] implemented a neat sensing technique here, integrating alternating magnets into the output ring gear which are monitored via a magnetic encoder. The wrist joint switches things up again, running via an inverted belt differential.

Running the show is an STM32 microcontroller, which talks to all the encoders, communicates with a Raspberry Pi over CAN bus, and handles all the necessary PID control loops and step generation for the drive motors. The plan is to run higher-level control on the Raspberry Pi which will run a ROS 2-based software stack. Already, the various joints look smooth and impressive in motion.

If you’re looking to learn about robot arms, you really can’t beat building one. We’ve featured a few projects along these lines before. Most of them aren’t exactly production-line ready, but they will teach you a ton about control, motion planning, and all sorts of associated skills. That experience can be invaluable if you intend to work with robots in industry.

Continue reading “3D Printed Robot Arm Built For Learning Purposes”

Let Hauntimator Steer Your Next Animatronic Display

Animatronic displays aren’t just for Halloween, and hackers today have incredible access to effective, affordable parts with which to make spectacles of light, sound, and movement. But the hardware is only half the battle. Getting everything synchronized properly can be a daunting task, so get a head start on your next holiday display with the Hauntimator by [1031-Systems].

Synchronizing control channels to audio is at the heart of solid animations.

After all, synchronizing movements, sound, and light by trial and error can get tiresome even in small setups. Anyone who makes such a display — and contemplates doing it twice — tends to quickly look into making things modular.

At its heart, Hauntimator works with a Raspberry Pi Pico-based controller board. The GUI makes it easy to create control channels for different hardware (for example, doing things like moving servos) and synchronize them to audio. Once an animation is validated, it gets uploaded to the control board where it runs itself. It’s open-source and designed to make plugins easy, so give it a look. There’s a video channel with some demonstrations of the tools that should fill in any blanks.

Intrigued by animatronics, but not sure where to begin? Get inspired by checking out this DIY set of servo-driven eyes, and see for yourself the benefits of smooth motor control for generating lifelike motion.

Robot Looks Exactly Like A Roll Of Filament, If Filament Had Eyes

[Matt Denton]’s SpoolBot is a surprisingly agile remote-controlled robot that doesn’t just repurpose filament spool leftovers. It looks exactly like a 2 kg spool of filament; that’s real filament wound around the outside of the drum. In fact, Spoolie the SpoolBot looks so much like the real thing that [Matt] designed a googly-eye add-on, because the robot is so easily misplaced.

The robot’s mass rotates around a central hub in order to move forward or back.

SpoolBot works by rotating its mass around the central hub, which causes it to roll forward or back. Steering is accomplished by tank-style turning of the independent spool ends. While conceptually simple, quite a bit of work is necessary to ensure SpoolBot rolls true, and doesn’t loop itself around inside the shell during maneuvers. Doing that means sensors, and software work.

To that end, a couple of rotary encoders complement the gearmotors and an IMU takes care of overall positional sensing while an ESP32 runs the show. The power supply uses NiMH battery packs, in part for their added weight. Since SpoolBot works by shifting its internal mass, heavier batteries are more effective.

The receiver is a standard RC PWM receiver which means any RC transmitter can be used, but [Matt] shows off a slick one-handed model that not only works well with SpoolBot but tucks neatly into the middle of the spool for storage. Just in case SpoolBot was not hard enough to spot among other filament rolls, we imagine.

The googly-eye add-on solves that, however. They clip to the central hub and so always show “forward” for the robot. They do add quite a bit of personality, as well as a visual indication of the internals’ position relative to the outside.

The GitHub repository and Printables page have all the design files, and the video (embedded just below) shows every piece of the internals.

The kind of hardware available nowadays makes self-balancing devices much more practical and accessible than they ever have been. Really, SpoolBot has quite a lot in common with other self-balancing robots and self-balancing electric vehicles (which are really just larger, ridable self-balancing robots) so there’s plenty of room for experimentation no matter one’s budget or skill level.

Continue reading “Robot Looks Exactly Like A Roll Of Filament, If Filament Had Eyes”

Print-in-Place Gripper Does It With A Single Motor

[XYZAiden]’s concept for a flexible robotic gripper might be a few years old, but if anything it’s even more accessible now than when he first prototyped it. It uses only a single motor and requires no complex mechanical assembly, and nowadays 3D printing with flexible filament has only gotten easier and more reliable.

The four-armed gripper you see here prints as a single piece, and is cable-driven with a single metal-geared servo powering the assembly. Each arm has a nylon string threaded through it so when the servo turns, it pulls each string which in turn makes each arm curl inward, closing the grip. Because of the way the gripper is made, releasing only requires relaxing the cables; an arm’s natural state is to fall open.

The main downside is that the servo and cables are working at a mechanical disadvantage, so the grip won’t be particularly strong. But for lightweight, irregular objects, this could be a feature rather than a bug.

The biggest advantage is that it’s extremely low-cost, and simple to both build and use. If one has access to a 3D printer and can make a servo rotate, raiding a junk bin could probably yield everything else.

DIY robotic gripper designs come in all sorts of variations. For example, this “jamming” bean-bag style gripper does an amazing, high-strength job of latching onto irregular objects without squashing them in the process. And here’s one built around grippy measuring tape, capable of surprising dexterity.

Continue reading “Print-in-Place Gripper Does It With A Single Motor”

Dual-Arm Mobile Bot Built On IKEA Cart Costs Hundreds, Not Thousands

There are many incredible open-source robotic arm projects out there, but there’s a dearth of affordable, stable, and mobile robotic platforms with arms. That’s where XLeRobot comes in. It builds on the fantastic LeRobot framework to make a unit that can be trained for autonomous tasks via machine learning, as well as operated remotely.

XLeRobot, designed by [Vector Wang], has a pretty clever design that makes optimal use of easy to obtain parts. In addition to the mostly 3D-printed hardware, it uses an IKEA cart with stacked bin-like shelves as its main frame.

The top bin holds dual arms and a central stalk with a “head”. There’s still room left in that top bin, a handy feature that gives the robot a place to stow or carry objects.

The bottom of the cart gets the three-wheeled motion unit. Three omnidirectional wheels provide a stable base while also allowing the robot to propel itself in any direction and turn on a dime. The motion unit bolts to the bottom, but because the IKEA cart’s shelf bottoms are a metal mesh, no drilling is required.

It’s all very tidy, and results in a mobile robotics platform that is cheap enough for most hobbyists to afford, while being big enough to navigate indoor environments and do useful tasks.

Continue reading “Dual-Arm Mobile Bot Built On IKEA Cart Costs Hundreds, Not Thousands”

Robotic Canoe Puts Robot Arms To Work

Most robots get around with tracks or wheels, but [Dave] had something different in mind. Sufficiently unbothered by the prospect of mixing electronics and water, [Dave] augmented a canoe with twin, paddle-bearing robotic arms to bring to life a concept he had: the RowboBoat. The result? A canoe that can paddle itself with robotic arms, leaving the operator free to take a deep breath, sit back, and concentrate on not capsizing.

There are a couple of things we really like about this build, one of which is the tidiness of the robotic platform that non-destructively attaches to the canoe itself with custom brackets. A combination of aluminum extrusion and custom brackets, [Dave] designed it with the help of 3D scanning the canoe as a design aid. A canoe, after all, has nary a straight edge nor a right angle in sight. Being able to pull a 3D model into CAD helps immensely in such cases; we have also seen this technique used in refitting a van into an off-grid camper.

The other thing we like is the way that [Dave] drives the arms. The two PiPER robotic arms are driven with ROS, the Robot Operating System on a nearby Jetson Orin Nano SBC. The clever part is the way [Dave] observed that padding and steering a canoe has a lot in common with a differential drive, which is akin to how a tank works. And so, for propulsion, ROS simply treats the paddle-bearing arms as though they were wheels in a differential drive. The arms don’t seem to mind a little water, and the rest of the electronics are protected by a pair of firmly-crossed fingers.

The canoe steers by joystick, but being driven by ROS it could be made autonomous with a little more work. [Dave] has his configuration and code for RowboBoat up on GitHub should anyone wish to take a closer look. Watch it in action in the video, embedded below.

Continue reading “Robotic Canoe Puts Robot Arms To Work”