All News | Boards | Chips | Devices | Software | Archive | About | Contact | Subscribe
Follow LinuxGizmos:
Twitter Facebook Pinterest RSS feed
*   get email updates   *

Raspberry Pi 4 based robot targets manipulation and service tasks

Jan 6, 2020 — by Eric Brown 1,625 views

Pollen Robotics is launching a modular, open source “Reachy” robot that runs Raspbian and Luos on a Raspberry Pi 4 with Google’s Coral AI Accelerator. The torso-like bot has one or two articulated, 7-DoF arms with pincers and a ball-joint head with dual cams.

French robotics firm Pollen Robotics has opened limited pre-orders for an open-spec robot for R&D, public hospitality and service, and basic manipulation duties. The human torso sized Reachy is built from modular, MCU-driven parts that communicate with each other using the Luos distributed OS (see farther below). Reachy is touted as “the only humanoid service robot that is open source and can manipulate objects.”

Reachy showing 7-DoF arm and painting skills in YouTube demo shots
(click images to enlarge)

Reachy can run AI algorithms including image and voice recognition routines and has self-learning capabilities. It has an expressive head with dual cameras that uses an “Orbita” ball-joint mounting technology developed by Pollen for multi-directional, 180-degree movement. The “bio-inspired.” 7 Degrees of Freedom (DoF) arms have 7x servo motors and can be fitted with pincers or human-like hands. The arms are said to be dexterous enough to enable prosthetic limb applications.

Reachy models
(click image to enlarge)

Reachy sells for about $9,000 with a single arm, $11,000 with the addition of the head, and $17,000 for a complete unit with two arms and a head. Only 15 units are currently available (or quite possibly “were” available”), due later this quarter.

The robot’s brain, which combines a quad -A72 Raspberry Pi 4 SBC with 2GB of RAM and a Google Coral AI accelerator, is located not in the head, but in the chest. This is also where you’ll also find a speaker and microphone.


The above computing details are derived from a hands-on preview from Tom’s Hardware. The report describes watching the robot pick up a series of blocks and play itself in a game of Tic Tac Toe. Additional feats are shown in the YouTube video farther below.

Reachy applications
(click image to enlarge)

The product will be available with open source hardware and software support, as well as forums for developers. For now, however, the details are rather sketchy. Pollen Robotics notes that the “TPU,” referring to Google’s Edge TPU (Tensor Processing Unit) AI chip, can run ML algorithms offline. It’s unclear if the Reachy uses the newly announced, solderable Coral Accelerator Module LGA module version of the Edge TPU or the earlier Coral USB Accelerator.

Reachy manipulating blocks (left) and exploded view of arm
(click images to enlarge)

The detail view above shows that the Raspberry Pi 4 connected HDMI, USB 3.0, and Gigabit Ethernet ports are exposed on the back of the unit. The two cameras are two Raspberry Pi cameras modified to have different fields of view. One is for examining the overall scene while the other is focused on manipulation tasks. This is likely the most advanced of the many available robots built around the Raspberry Pi, and the first we’ve seen based on the RPi 4.

The Orbita ball-joint technology for the head is accompanied by separate actuators that individually control the two antennas to express emotion. Presumably, they also amplify the RPi 4’s WiFi.

Reachy detail view (left) and Tom’s Hardware photo of internals, with box highlighting Raspberry Pi 4 with heatsink
(click images to enlarge)
Source: Tom’s Hardware (image on right)

The robot is powered by a 100-240V supply, and the arms can together lift weights of up to 500 grams. An alternative version is available for applications that “require longer-term usage and particular strength,” says the company.

Reachy runs Raspbian with ROS, backed up with an open source Python library. The Coral AI software running on the Edge TPU supports TensorFlow Lite. Also available are DialogFlow and the Snips AI voice control assistant.


Tom’s Hardware says the robot’s modules come from Luos Robotics, but we saw no indication that Luos is a hardware company. The OS that runs on them is Luos, however. Luos is billed as a ROS for microcontrollers and can run on MCUs with as little as “24ko,” which may refer to kilobits. Using RS485 technology, Luos devices can send messages at 10Mbps.

Inspired by “many different industrial-grade bus and distributed operating systems,” Luos “sorts application code and drivers into modules,” and then can detect, localize, and list all of them, says Luos. Once several nodes are connected, the rest can automatically self-organize into a single system image. Available with a C/C++ based API, Luos can translate to Json and can be accessed by high level languages including Python, Ruby, JavaScript, and ROS.

Reachy YouTube debut video

Further information

The Reachy robot sells for 7,990 to 15490 Euros (a bit less than $9,000 to $17,000), and is available for pre-order here. If you miss out on the first 15 models, which are due to ship in the current quarter, you can presumably be added to their waiting list for a larger production run. More information may be found on the Pollen Robotics website.

(advertise here)

Print Friendly, PDF & Email

One response to “Raspberry Pi 4 based robot targets manipulation and service tasks”

  1. Simon says:

    > “24ko,” which may refer to kilobits
    It’s 24kB, it’s been corrected on Luos website!

    Reachy uses Luos dev boards, some of them are available in Generation Robots’ website:

Please comment here...