Robotics resources

Here are some resources for robotics (software and hardware) that I have found useful and/or interesting. It's still incomplete but it's a good start.

Robotics software/simulation:

ROS. The "Robot Operating System": A bit overloaded name, as it's not really an operating system, but it's (an open source) middle layer to interface wth a lot of robots and sensors with a useful visualization tool (RVIZ).
MuJoCo: Multi-Joint dynamics with Contact, is a general purpose physics engine (acquired by Google's DeepMind) that is used for a lot of robot simulations, to generate positions, coordinates, and avoid joint violations.
Isaac Sim: Nvidia's PhysX engine. It's also used for robot simulations, multi-sensor RTX rendering (GPU), with support of multiple existing robots.

Recent AI Papers:

Not covering much on classical robotics like SLAM, VSLAM, Path Planning, Kinematics/Dynamics, Jacobian Matrices, CNNs, etc. But very helpful to have an understanding of them.

RT-1 (Robotic Transformer): A vision-language AI model for robot control that processes sequences of visual inputs alongside textual instructions through a transformer, generating specific motion commands that can drive robot actuators.
SayCan, Grounding Language in Robotic Affordances: a model (based on PaLM) to combine semantic knowledge with pretrained robot behaviors to enable robots to execute complex text instructions in real-world environments.
RT-2: a model that combines vision-language AI with robot control by expressing robot actions as text tokens, enabling better generalization and reasoning capabilities that emerge from internet-scale training.
Mobile Aloha: a low cost, a teleoperation robot that enables robots (Aloha Kits) to learn complex whole-body mobile manipulation tasks from human demonstrations, successfully performing activities household tasks through co-training with existing datasets.

Robot foundation models:

Nvidia Cosmos: a foundation model platform designed to help Physical AI/Robotics: a set of pre-trained Diffusion-based world foundation models for Text->Video and text+video->video and a set of pre-trained autoregressive-based world foundation models for Video2World generation. Models are trained on 20M hours of videos and the model weights range from 4B to 14B.
π0 from Physical Intelligence (π): Pi's first general-purpose robotic foundation model that could be fine-tuned to a diverse range of tasks (folding laundry, cleaning a table, scooping coffee beans, etc) and that could control a variety of different robot type (robots with two arms, single-arm platforms, mobile robots, etc).

Robot datasets:

Sadly, there aren't many large datasets for robots out there yet.

DROID Dataset: A diverse robot manipulation dataset with 76k demonstration trajectories (350h of interaction data), collected across 564 scenes and 86 tasks.
ARMBench (released by Amazon): contains over 190K objects in industrial product-sorting settings, great for pick and place robots.
AGIBot World: 1 million+ trajectories from 100 robots. 100+ 1:1 replicated real-life scenarios across 5 domains, great for single arm and mobile dual arms robots.
Open X-Embodiment (Datasets and RT-X Models): probably the largest os dataset to-date: it contains over 1 million real robot trajectories from 22 different robots (xARM, Franka, Sawyer, Kuka, etc), demonstrating 527 skills. RT-X models demonstrate that a single "generalist" policy can effectively transfer learning between different robots and tasks.

Robotics Hardware (that you can buy now):

Robot Arms:

xArm from uFactory: I have used these robot arms and they have worked great so far. Used in research and in some light industrial applications. Great cost too and good Python/ROS SDK. Available in 5, 6 and 7 DOF. Company is based in China.
Franka Emika: A 7 DOF arm mostly used for research. More expensive than the xARMs for with a few more functionalities. Good Python/ROS SDKs. Company is based in Germany.
Universal Robots (UR): Industry standard collaborative robot arms. Mostly used in production and industrial environments. More expensive than xARM and Franka. Available with multiple payloads and DOF. Python SDK. Company is based in Denmark.
Aloha Kits: Set of different Aloha arms (Aloha Solo, Aloha Stationary, Mobile Aloha) for teleoperation, data collection and model training. Only used in research.
Agile X Piper: A low cost ($2.5K) robot arm with 1.5Kg payload with 6DOF

Other robots:

Unitree Go2: Fantastic robot dog with great kinematics & hardware. Lidar and multiple cameras and sensors available. Starts at $1.6K. Not a big fan of their software Software/SDKs, but it has been improving.
Unitree B2: An industrial grade robot dog capable of bigger payloads (~40Kg) than the Go2. Larger batteries and longer range Lidar.
Unitree G1: Entry level, short-sized humanoid robot with a 3D lidar, depth camera, 23 DOF, moving speed of 2m/s. Starts at $16K.
Unitree H1: Full-sized humanoid robot (180cm). Speed of 3.3m. 3D lidar and depth camera (Intel RealSense D435i). Stronger motors and bigger payload than the G1.
Xiaomi Cyberdog: Another quadruped dog. It runs on NVIDIA's Jetson Xavier NX. The robot features 11 sensors, including binocular ultra-wide cameras, along with Intel's RealSense D450 depth module. It can walk at speeds up to 3.2 meters per second and carry loads weighing up to 3 KG.
LimX Dynamics W1: Wheeled quadruped Robot for all terrains.
LimX Dynamics CL-1: (no much data about it)
LimX Dynamics TRON1: Biped Robot, Python SDK and MuJuCo, Isaac Sim and Gazebo integration. Add-ons: wheels, radar, robot arm, camera and gimbal.
DeepRobotics: A very neat, recently released robot dog with great kinematics and wheels.
Igus Delta pick and place robot: a high-speed, low-cost delta picking robot.
Tita Biped robot: Another biped robot with wheels on the legs.

Some Robot parts:

Compute:

Nvidia Jetson Series: the Nvidia Jetson Orin Nano 4Gb I find it to small for anything serious. Probably best if it's 8Gb+.
Intel NUC computers (with Linux): they are great for doing general computing for robotics (no GPU). It drains about 3-5 amps of current depending on compute workload.
ODROID small computers: much better (small) computers than Raspberry Pis. Probably some of the most powerful computers for their size and power consumption. Always use an external fan for them.

Motor Controllers:

VESC Project: The VESC (Vedder Electronic Speed Controller) project is an open-source electronic speed controller for electric motors. It can do precise motor position control without using an encoder with "silent HFI" (High-Frequency Injection), which enables the VESC to determine the motor's position accurately without any external sensors/encoders by analyzing the response of the HFI to deduce the rotor position.
VESCular6: a multi-functional motor controller based on the VESC project which provides precise position control algorithms, Degree Per Second (DPS) speed control and Goto control. High-level commands can be issued through VESC-Tool and through ROS.
ODrive: high power (3kW Continuous Power, 5kW peak), 14-58V range, with CAN, UART, Step/Direction and GPIO. Works with Python, Arduino, CAN, and ROS2.
Obot controller: An open source controller that works via USB with Linux, ideally for Nvidia Jetson Nano, Raspberry Pi, etc. Originally desgined fot the Obot robot.
Solo Motor Controller: Supports DC, BLDC, PMSM, and ACIM Motors, Sensored and Sensorless. Torque, Speed Control and Position Control. Library for C++, Python, Arduino and Mathlab.

Lidars:

Velodyne: A Californian company and the OG in the space (started as a lidar for the DARPA Grand Challenge) and in 2023 it merged with Ouster. I used the Velodyne VLP16 a lot for robots, but got replaced by the new OS series by Ouster.
SICK Lidars: A German manufacturer; very reliable and widely used in industrial applications. The TIM551 (2D) is a common one for indoor robotics.
Avia Lidar: A $1.6K 70+ degree lidar with 450 mts max of range, 2cm precision and IP67 rating.
RoboSense Lidars: 3D options: Airy, Helios, Bpearl.
Hokuyo Lidars: A Japanese manufacturer of 2D lidars, also popular in indoor robotics.
Unitree L2 3D Lidar: A ~$419USD 3D Lidar.
DJI Lidar: Normally used for cameras, but could be used for robotics.

Depth Sensors:

Intel Realsense: Widely used in robot applications. A few different models are available with good software support.
ZED 2: ZED 2 by StereoLabs. A good alternative to Realsense.
Orbbec: Many options, from Stereo vision, TOF, and Structured Light Cameras.
Microsoft Kinect: Sadly Microsoft killed the Kinect (again), but still a good one.

Actuators:

There are many so many options for motors. Here some popular ones:

MyActuator motors: these are cheap, powerful brushless motors with positional encoders, and many communication buses (CAN, RS485, EtherCAT). They have replaced in many cases harmonic drives.
Robstride: another cost efficient powerful brushless motors with positional encoders.
Xiaomi CyberDog motor: Great small motor with relatively high torque used in the Xiaomi robot CyberDog. Great price.
DeepRobotics J60, J80, J100: These are some of the motors used in the DeepRobotics dogs and humanoids with CAN Bus.
Dynamixel P and H Series: These are somewhat high-end servos, a bit pricey.

Others:

Controllers (useful for local teleop):

Normally these remote controllers are use for other purposes (Gaming, RC planes, etc) but are very useful for robotics:

Long range:

Short range (Bluetooth):

Any of the XBox/gaming Bluetooth controllers

Electronics stores, and component stores:

Mouser: A US based Integrated Circuit (IC) supplier. Most ICs are available here. Mouser and Digikey are the leaders in providing from one IC to millions in reels for pick and place machines.
DigiKey: A US based Integrated Circuit (IC) supplier. Most ICs are available here. Mouser and Digikey are the leaders in providing from one IC to millions in reels for pick and place machines.
Pololu: US based robotics and electronics store with some great custom made boards like DC converters, motor drivers, bus communication boards, etc.
SparkFun Electronics: US based electronics store with multiple custom mde boards from GPS, communication boards, Arduino Shields, ESP32 shields, audio boards, etc. Some of their designs are open source.
Adafruit: US based DIY electronic store. Many custom made boars for LEDs, communication, Arduino shields, etc. Most of their designs are open source.
SeedStudio: Shenzhen-based electronics store. They have different designs than Pololu/Adafruit/Sparkfun, and have some great custom made alternatives.
JLCPCB Parts: This is like a Chinese Digikey/Mouser. If you also make the PCBs with them, they will assemble it at a low cost (and fast).
McMaster-Carr: This US-based store has basically every hardware component. From the smallest screws to welding machines. If budget is not a problem, you can get most of your hardware components here. Their catalog is phenomenal and it includes 3D CAD step files for almost every part.
Vevor: This an affordable generic brand that produces a large number of tools and equipment.
RobotDigg: This is a China-based robot components store that has a large number of different industrial and robot components. From motors to CNC, tools and drivers.
SenRing: Some very cool slip rings and rotary unions for 360 degree joints.
Moflon: Same, very cools 360 degree slip rings; HDMI and Ethernet, high-power, pneumatics, etc.
Industrial mecanum wheels: Industrial mecanum wheels with gearboxes.
Piab end-effectors: Multiple end-effectors for robot arms. Suction, soft grippers, vacuum pumps, etc.

For a robotics lab (some tools I use):

JBC Soldering iron: I have tested many soldering irons (Weller, Hakko, Metcal, etc), and this is by far my favorite.
Owon 200Mhz Oscilloscope: these are great low-cost oscilloscopes. I have an older version of this one, but this newer one has also a waveform generator.
Riden Programmable Power Supplies: These are low-cost, high-power programmable (switching) power supplies (multiple versions available). I have multiple versions of these (RD6006, RD6012, RD6024) and they work fantastic. Note that for some applications, these might be a bit noisy due to the switching power supply (specially with no/low loads). For my applications they work great.
A Linear Power Supply: If you still want a low noise linear (transformer-based) power supply, I have used this one and it works great.
Saleae Logic Analyzer Pro 8: I have loved this logic analyzer for many years. The software is also pretty neat.
Sensepeek PCBite Probes: These are some great miniature "hands-free" PCB probes, including high frequency probes and magnetic PCB holders.
Bambu Lab 3D Printer: A great Fused Deposition Modeling (FDM) 3D printer. I have used many of the older 3D printers, and it was always a mess. This one "just works".
A Programmable DC load: A used a very similar version of this generic one. A low-cost DC programmable load can be very helpful in a lab to simulate motor loads, batteries, or any DC load.
A 3D mouse for CAD/PCB Design: I have the 3D space mouse (but you could use any od their 3D mouse options; they are all great) and it's a huge boost in CAD and PCB design productivity for me. I used it with Catia, SolidWorks and Altium. There's a bit of a learning curve, but later you can't work without it.
Bernstein tools: High quality German tools for electronics.
iFixit tools: iFixit, the popular brand for phone repairs, has a great set of custom tools, great for a robotics lab.

Updated on: