CtrlK
TurtleBot is an open source, low cost mobile robot platform built around the Robot Operating System (ROS) and intended primarily for robotics education and research. Created in late 2010 at Willow Garage by Tully Foote and Melonee Wise, the platform has gone through four major generations, each replacing the previous mobile base, sensors, and onboard computer while keeping the same core idea: a small, affordable, hackable robot that runs the same software stack used on much more expensive research machines. Since 2017 hardware manufacturing has been led by ROBOTIS of South Korea, with software and community efforts coordinated by Open Robotics (the Open Source Robotics Foundation, OSRF). For TurtleBot 4, Clearpath Robotics joined as a third partner.
The original TurtleBot was a Thanksgiving 2010 hack: take an iRobot Create base (a developer version of the Roomba vacuum), bolt a Microsoft Kinect on top, plug in a netbook, and ship the whole thing as a kit so that grad students and self taught roboticists could run ROS without buying a $400,000 PR2. That decision shaped a generation of mobile robotics. Today TurtleBot is the standard hardware on which most ROS tutorials are written, the platform behind dozens of university courses, and the reference robot for benchmarking SLAM, navigation, and increasingly large language model driven robotics demos.
Willow Garage was a robotics research lab in Menlo Park, California that funded the early development of ROS and built the PR2, a two armed mobile manipulator that became the unofficial flagship for the new framework. The PR2 was extraordinary, and extraordinarily expensive. By 2010 the lab was getting a steady drumbeat of complaints from the ROS community: the software only seemed to run well on robots no normal lab could afford.
Melonee Wise and Tully Foote, both engineers at Willow Garage, had been kicking around ideas for a cheap entry level robot. The Microsoft Kinect launched on November 4, 2010, and a hobbyist driver appeared a few days later. Around Thanksgiving Wise grabbed a spare iRobot Create, screwed a Kinect to a wooden board on top, plugged in an Asus 1215N netbook, and demonstrated that the whole rig could run the existing ROS navigation stack. The total bill of materials sat under $1,200, an order of magnitude cheaper than anything else in the lab. The robot was called TurtleBot, a nod to Turtlesim, the toy turtle simulator inside ROS used to teach beginners. Wise summed up the naming logic in a later interview: "Everything in ROS is turtles."
Willow Garage announced the platform in 2011, released the design as open source, and arranged for partners to sell complete kits. The bet paid off. Within a year TurtleBot had become the canonical way to learn ROS, partly because the hardware was cheap enough that universities could buy a fleet, and partly because every ROS tutorial author standardized on it.
The first generation TurtleBot, often written TurtleBot1 or TB1, was released through partner vendors in 2011 with a list price near $1,200 for the assembled kit. The design was deliberately scrappy.
| Component | Specification |
|---|---|
| Mobile base | iRobot Create (Roomba developer variant) |
| Depth sensor | Microsoft Kinect (structured light) |
| Onboard computer | Asus Eee PC 1215N netbook (Atom dual core, Nvidia ION) |
| Battery | 3,000 mAh pack supplying the Kinect and laptop |
| Power board | Custom TurtleBot board with single axis gyro |
| Mounting | Acrylic stack plates with standoffs for additional sensors |
| Top speed | ~0.5 m/s (limited by Create base) |
The Create base could not move very fast, the Kinect needed at least 0.5 m of clear space ahead to return useful depth data, and the netbook had no GPU worth using. None of that mattered. The kit ran the same move_base navigation stack, the same gmapping SLAM library, and the same rviz visualizer that students saw demonstrated on the PR2. For most newcomers it was the first robot they ever wrote real code for.
The original boxed kits were sold by I Heart Engineering and later by Clearpath Robotics. A community variant called the TurtleBot Roomba Edition swapped the Create for a hacked Roomba 500 series.
TurtleBot 2 launched in October 2012. The biggest change was the base. iRobot had announced it was winding down the original Create, so the TurtleBot team partnered with Yujin Robot of South Korea to use a new mobile platform called Kobuki, designed from the start with research robotics in mind.
| Component | Specification |
|---|---|
| Mobile base | Yujin Robot Kobuki |
| Depth sensor | Kinect, with Asus Xtion PRO LIVE and Orbbec Astra options later |
| Onboard computer | Asus 1215N netbook (later swapped for Intel NUC and other PCs) |
| Battery | 2,200 mAh (small pack) or 4,400 mAh (large pack) |
| Charging | Auto docking via the Kobuki dock |
| Bumpers | Three front bump sensors integrated into Kobuki |
| Top speed | 0.65 m/s |
Kobuki had a real wheel encoder, a factory installed gyro, an auto charging dock, and a bumper array. It also had a documented serial protocol with C++ and Python drivers. From a software point of view, TurtleBot 2 was the first generation that felt purpose built for research instead of cobbled together from consumer parts.
TB2 also picked up a wide ecosystem of optional sensors and arms. The Kobuki base accepted up to 5 kg on its top plate, which left room for a lightweight manipulator like the PhantomX Pincher arm or a small RPLidar unit added on top. Many labs ran TurtleBot 2 with a LIDAR scanner instead of, or alongside, the Kinect, particularly once Microsoft discontinued the Kinect for Windows in 2015 and the Kinect for Xbox One stopped being a comfortable fit.
The second generation also gained a serious Gazebo simulation model. For the first time you could run a TurtleBot in simulation that closely matched the real hardware, which made it easier to teach robotics in courses where you only had three physical robots for thirty students.
By 2016 the original TurtleBot story needed a sequel. Yujin Robot had stopped making the Kobuki in volume, the Kinect was effectively dead, and netbooks had vanished from the market. At ROSCon 2016 in Seoul, the Open Source Robotics Foundation and ROBOTIS announced TurtleBot 3, a clean sheet redesign manufactured and distributed by ROBOTIS. The robot officially launched in May 2017 at ICRA in Singapore, with Intel as a launch partner.
TB3 was much smaller than TB2, used a 360 degree LIDAR instead of a depth camera as the primary sensor, and replaced the laptop with a single board computer. It came in three flavors.
The Burger is the entry level model: tall and narrow, two wheeled, designed to be the cheapest viable robot that still runs the full SLAM and navigation stack. The Waffle and Waffle Pi are wider, square shouldered platforms with extra payload capacity, an onboard camera, and the option to carry a small arm or a heavier compute board. The Waffle was originally sold with an Intel Joule 570x SBC and an Intel RealSense R200 depth camera; when Intel cancelled the Joule line in 2017, the Waffle Pi replaced it with a Raspberry Pi 3 and a Pi camera. The Joule based Waffle was eventually discontinued, and current Burger and Waffle Pi kits ship with Raspberry Pi 4.
| Specification | Burger | Waffle Pi |
|---|---|---|
| Footprint (L x W x H) | 138 x 178 x 192 mm | 281 x 306 x 141 mm |
| Weight | 1.0 kg | 1.8 kg |
| Wheel motors | 2 x DYNAMIXEL XL430-W250 | 2 x DYNAMIXEL XM430-W210 |
| Maximum linear speed | 0.22 m/s | 0.26 m/s |
| Maximum angular speed | 2.84 rad/s | 1.82 rad/s |
| Maximum payload | 15 kg | 30 kg |
| LIDAR | LDS-01, later LDS-02 (360 deg, ~3.5 m range) | LDS-01, later LDS-02 |
| IMU | 3 axis gyro + 3 axis accelerometer | Same |
| Camera | None (optional) | Raspberry Pi Camera Module v2 |
| Microcontroller | OpenCR 1.0 (ARM Cortex M7 at 216 MHz) | Same |
| Single board computer | Raspberry Pi 4 (current shipping) | Raspberry Pi 4 |
| Battery | 11.1 V 1,800 mAh LiPo | 11.1 V 1,800 mAh LiPo |
| Operating time | ~2.5 h | ~2 h |
| Launch price (2017) | $549 | $1,799 (Waffle), $1,399 (Waffle Pi) |
The motors are the part that made TB3 different from a hobbyist robot. ROBOTIS makes the DYNAMIXEL line of smart serial servos, which expose position, velocity, and current control over a daisy chained RS485 or TTL bus. Each wheel has its own DYNAMIXEL with its own encoder, and each motor reports back its load and temperature. That means the same servos used in research arms drive the wheels, which makes the robot quieter, more accurate, and far more programmable than a typical hobby base. The motors also slot into the same modular ROBOTIS frame system, so users can rebuild a Burger into a four wheeled cart, a tracked rover, or a small manipulator without machining new parts.
The controller board, OpenCR (Open source Control module for ROS), runs an Arduino compatible firmware, talks to the DYNAMIXELs and the IMU, and acts as a bridge between the SBC running ROS and the low level hardware. Sample code, schematics, and PCB layout are all open source.
The Burger has held up as the workhorse model. Its small footprint fits comfortably in a classroom, the LIDAR returns enough data to do real SLAM, and the price point still lets a department buy a fleet of fifteen for under $10,000. The Waffle Pi has been the platform of choice when a class needs a camera in addition to a LIDAR, or when students want to add a small servo arm.
At ROSCon 2021 Open Robotics, Clearpath Robotics, and iRobot announced TurtleBot 4. The new robot launched on May 4, 2022, with shipping starting in July 2022. The leap this time was a full move to ROS 2, with the platform officially supported on Galactic and Humble at launch and on Iron and Jazzy in later updates.
TB4 ditches the ROBOTIS chassis for a return to an iRobot base, this time the new Create 3, which is essentially a Roomba i3 with a research firmware that exposes ROS 2 topics directly over the iRobot side of the stack. The robot ships in two configurations.
| Specification | TurtleBot 4 Lite | TurtleBot 4 Standard |
|---|---|---|
| Mobile base | iRobot Create 3 | iRobot Create 3 |
| Footprint (L x W x H) | 342 x 339 x 192 mm | 342 x 339 x 351 mm |
| Weight | 3.27 kg | 3.95 kg |
| Maximum linear speed | 0.31 m/s (safe), 0.46 m/s (unrestricted) | Same |
| Maximum angular speed | 1.90 rad/s | Same |
| Maximum payload | 9 kg (15 kg with custom mods) | Same |
| LIDAR | RPLIDAR A1M8 (360 deg, 12 m range) | Same |
| Camera | OAK D Lite stereo + RGB | OAK D Pro stereo + RGB with IR projector |
| Onboard computer | Raspberry Pi 4B (4 GB) | Raspberry Pi 4B (4 GB) |
| OLED display | No | Yes (128 x 64) |
| User buttons | None | 4 programmable |
| User power rails | 5 V, 3.3 V, VBAT | 3.3 V, 5 V, 12 V, VBAT |
| Battery | iRobot 26 Wh Li-ion | Same |
| Operating time | 2.5 to 4 h | 2.5 to 4 h |
| Launch price (2022) | $1,195 | $1,850 |
The sensor change is the interesting one. TB3 used a fairly cheap 2D LIDAR and treated cameras as optional. TB4 ships with both a fast 12 m 2D LIDAR and an OAK D, the open source RGB-D and machine vision module from Luxonis. The OAK D has its own onboard inference accelerator (Movidius VPU) so neural network workloads such as object detection or pose estimation can run at the camera, leaving the Raspberry Pi free to handle ROS messaging and navigation. The Pro version on the Standard kit adds an IR dot projector for better depth in low texture environments.
Clearpath Robotics took over manufacturing and global distribution; the company was later acquired by Rockwell Automation in November 2023, which inherited TurtleBot manufacturing duties.
Every TurtleBot generation ships with software released under permissive open source licenses, mostly Apache 2.0 and BSD-3, with documentation under CC BY 4.0. The exact software stack tracks whichever ROS distribution is current at the time the robot launches.
For the early TurtleBots, the canonical stack was ROS Electric, then Fuerte, then Hydro, with move_base for path planning, gmapping and later cartographer for Simultaneous Localization and Mapping, and amcl for localization. TurtleBot 3 was launched with ROS Kinetic and continues to be supported through Noetic on the ROS 1 side and through Humble on ROS 2. TurtleBot 4 was designed for ROS 2 from the start, using Nav2 (the ROS 2 navigation stack) and SLAM Toolbox or Cartographer for mapping.
Most TurtleBot users never assemble the full stack from source. ROBOTIS and Open Robotics maintain pre built Debian packages, Docker images, and a Raspberry Pi SD card image with Ubuntu plus ROS already installed. A typical first session looks like flashing the SD card, joining the robot to the lab Wi-Fi, sourcing the ROS environment, and running ros2 launch turtlebot4_navigation slam.launch.py.
Gazebo simulation models exist for every generation, and TurtleBot 3 and 4 also have official models in NVIDIA Isaac Sim. The simulators match the real hardware closely enough that students often develop and debug in simulation, then move to the real robot only at the end. That workflow has become the default in most graduate level robotics courses.
For multi robot research, TurtleBot is also a common test fleet for swarm SLAM, distributed planning, and human robot interaction studies. Twenty Burgers fit comfortably in a single lab.
TurtleBot is, fairly or not, the implicit default robot in most modern robotics education. A non exhaustive list of well known courses and textbooks that use it:
There is also a research literature centered on TurtleBot itself: a 2019 paper, Turtlebot 3 as a Robotics Education Platform (Amsters and Slaets, in Robotics in Education), evaluated outcomes of using Burgers in master's level courses at KU Leuven, and concluded that students who started with TB3 picked up ROS, SLAM, and Nav2 in roughly half the time of students starting from a self assembled robot.
In AI specifically, TurtleBot has become a frequent demonstration target for two kinds of research. The first is reinforcement learning for navigation, where the robot's standardized observation and action spaces make it easy to compare PPO, SAC, and similar agents across labs. The second, more recent, is large language model driven robotics, including "talk to your robot" demos using Code as Policies, SayCan, or VoxPoser style stacks where a language model issues high level commands and TurtleBot's Nav2 stack handles the low level execution. The platform is too underpowered for serious foundation model inference on board, but it is exactly the right size for end to end demos when paired with a workstation.
The modular hardware and open documentation have generated a long tail of third party accessories.
ROBOTIS also sells a series of fully assembled "friend" variants of TurtleBot 3 with the platform reshaped into a tank, a forklift, a delta, or a bipedal walker for specialized course assignments.
TurtleBot is not the only mobile robot used in research. The closest comparisons are listed below, with rough 2025 prices and intended audience.
| Platform | Maker | Base | Typical sensors | Compute | Approx. price | Niche |
|---|---|---|---|---|---|---|
| TurtleBot 4 Lite | Clearpath / iRobot | Create 3 | RPLIDAR A1, OAK D Lite | Raspberry Pi 4 | $1,200 | Education, ROS 2 reference robot |
| TurtleBot 3 Burger | ROBOTIS | DYNAMIXEL diff drive | LDS-02 LIDAR | Raspberry Pi 4 | $700 | Cheapest serious ROS robot |
| Pioneer 3-DX | Adept / Omron | Custom diff drive | SICK LMS LIDAR optional | External PC | $4,000+ | Older lab standard, rugged |
| Clearpath Husky A200 | Clearpath / Rockwell | 4WD skid steer, 75 kg payload | Customer choice | Customer choice | $25,000+ | Outdoor, heavy payload research |
| Clearpath Jackal | Clearpath / Rockwell | 4WD skid steer, 20 kg payload | Customer choice | Customer choice | $13,000+ | Faster outdoor research |
| NVIDIA JetBot | NVIDIA reference design | DC motors, custom | Pi camera | Jetson Nano | $250 (DIY) | Tiny deep learning education robot |
| Roomba based DIY | Various | iRobot Create | Customer choice | Customer choice | Variable | Hobbyist ROS robots |
Pioneer 3-DX deserves a footnote because it was the dominant academic mobile robot before TurtleBot existed; many older robotics papers were written on Pioneers running ARIA or early ROS. The Husky and Jackal occupy the rung above TurtleBot, where research moves outdoors and payload starts to matter. JetBot occupies the rung below: it is closer to a Raspberry Pi car than a research robot, but it is the right pick if the entire course is about deep learning rather than classical SLAM.
Within that landscape TurtleBot has a stable position. It is the robot you buy when you want a mobile platform that runs the same code as a much more expensive lab robot, costs less than a high end laptop, and has an existing curriculum already written for it.