For autonomous robotics you need the serious number crunching ability of a GPU (graphics processing unit). There are only 2 main vendors of GPUs: AMD and Nvidia
Only Nvidia has “embedded” GPU / CPU combos packing all their computing power into packages that are small and efficient enough for autonomous robots.
Nvidia’s Isaac software for autonomous robots is free, complete with detailed hardware assembly instructions. But you can’t buy robots from Nvidia
Sharx Robotics is the only vendor offering the advanced Nvidia robotics reference designs fully assembled and ready to go!
Jetbot: A GPU on wheels
Jetbot is a powerful Nvidia® GPU with battery, Wifi and wide angle high-res camera, mounted on a 3D printed chassis and able to move around using a pair of toy motors with a very basic motor controller. The samples and tutorials for Jetbot are shown in Python with Jupyter notebooks. Does not support Isaac.
This is the robot to get if you have budget constraints or want to primarily work on GPU algorithms or AI models with only a camera input, and where movement of the robot is only guided by the camera or a video game controller.
Kaya: A small autonomous robot with the essential features for research
Kaya RS is a powerful Nvidia® GPU with battery, IMU, Wifi and Intel Realsense stereo depth camera, mounted on a 3D printed holonomic drive chassis and able to move around using 3 Robotis® brand precision servo motors with integrated servo controllers and motion feedback. Programmed with Nvidia’s Isaac SDK.
Kaya’s Realsense depth camera gives the robot an awareness of how far it is from obstacles, and the acceleration / rotation sensors in the IMU (inertial motion unit) together with the feedback from the wheels gives it an awareness of its own movement. This allows Kaya to create maps and plot paths to goals using a subset of Nvidia’s full Isaac robotics software suite.
Carter: The full size, standard robot for Nvidia’s autonomous robotics software suite
Carter is the reference standard. It runs the full Nvidia Isaac robotics software suite for autonomous robotics research. This allows mapping, localization, Lidar SLAM, local and global path planning, obstacle avoidance, tag following, delivery, and many other important robotics software tasks and building blocks. It is fascinating how accurately this real Carter robot and its “digital twin” simulation in Nvidia’s Omniverse virtual world track each other, and really teaches Nvidia’s “Sim-to-Real” development methodology. For corporate warehouse logistics planning this is the robot that Nvidia’s Cu-Opt/Re-Opt uses as an example
ReachR™: A Carter robot with industry standard robot arm
Carter can go places and observe but can’t manipulate its environment. For that, Sharx Robotics is developing the ReachR™ robot which is essentially a Carter together with a Universal Robots UR3e arm on an enhanced mobile platform.
Lidars as used by Nvidia’s Isaac software and Sharx Robotics hardware use up to 16 beams of laser light continuously scanned 360 degrees around the robot. This allows the robot to generate a precision map of its surroundings, then compare its current location to points on the map, and plan a path while dynamically avoiding obstacles. Due to their standard-setting reliability as well as free availability of matching driver software, our preferred supplier of Lidars is Velodyne (as of 2023 a division of Ouster).