Which Pi It Actually Supports

This is the single biggest gotcha. Kits designed around the Pi 4 may not have a mounting pattern, camera cable or power headroom for a Pi 5, and the Pi 5 uses a different, smaller camera connector, so an older kit’s ribbon cable will not fit. Kits built for a Pi Zero 2 W save weight and battery but give you far less compute for OpenCV. Read the compatibility line on the listing, then check the vendor’s current documentation, not an archived version.

Camera and Pan-Tilt Quality

The camera is the reason you chose a Pi, so do not treat it as a bonus. Look for a pan-tilt or 2-DOF gimbal, because a fixed camera makes target tracking almost pointless. Also look at how the module is mounted: several kits rivet the camera in place, which means a failed module is not user replaceable. Budget kits ship low megapixel modules that need bright light, and swapping in a better Pi camera later is usually cheap and worth it.

Servo Count and Drive Type

Servo count is a proxy for how hard the software will be. Two drive motors and one camera servo is a weekend. Ackermann steering changes the kinematics in a useful way. Mecanum wheels need a proper inverse kinematics matrix before the robot moves sideways correctly. Twelve or eighteen servos on a quadruped or hexapod means gait tables, interpolation and leg IK, which is a much bigger and much more rewarding project.

Power, Batteries and the HAT

A Pi 4 or Pi 5 driving motors and servos is a real power problem, and brownouts that reboot the Pi mid run are the classic symptom. Check whether the kit includes a proper robot HAT with a regulated 5 V supply and separate motor rail, whether 18650 cells and a charger are in the box, and whether you can charge without disassembly. Kits that ship no battery at all can add a meaningful amount to the real cost.

Documentation and Open Source Code

On a Pi robot, software support matters more than plastic quality. The best vendors publish versioned online docs and a public GitHub repository you can read, fork and fix. The worst ship a download link on the box and a zip file that assumes an OS release from three years ago. Before you buy, open the vendor’s docs site and the repo, check the last commit date, and see whether the install steps name a specific Raspberry Pi OS version.

The TurboPi is the kit to buy when the robot is the smallest part of the project and the perception stack is the point. It rides on mecanum wheels bolted to a metal chassis, which means genuine 360 degree omnidirectional movement and a real inverse kinematics exercise before it will strafe cleanly. Hiwonder builds the software around Python, OpenCV and ROS 2, so this is the natural pick if you are heading towards navigation stacks, SLAM experiments or a robotics course that expects ROS. The two axis HD camera feeds FPV video and drives the built in colour recognition, object tracking, line following and autonomous driving demos, and recent bundles add multimodal AI model and voice interaction support. The catch is the listing itself: TurboPi is sold in a confusing spread of variants, standard and advanced, with a Raspberry Pi 5 included or not, at very different prices, and reviewer numbers on each individual listing are small. Read the bundle title carefully before you order.

The PiDog is where a Raspberry Pi robot stops being a car with a camera and becomes a genuine robotics problem. Twelve servos across four legs mean you are coordinating a gait, not setting two PWM values, and SunFounder ships around 32 prebuilt actions, walking, sitting, standing, head shakes and tail wags, that you can read as Python and then take apart to understand the interpolation and leg inverse kinematics behind them. The sensor suite is the most complete here: a camera, a six axis IMU for balance, an ultrasonic range finder, a touch sensor, a sound direction module, a speaker and a microphone, which together are enough for balance experiments, reactive behaviours and voice interaction. It runs on the Pi 5, 4B, 3B+, 3B or Zero 2 W, and the board is not included, though a bundle with a Pi Zero 2 W and SD card exists. Assembly is long and fiddly, and twelve cheap servos will eventually need a spare or two.

Six legs with three servos each gives the Big Hexapod eighteen degrees of freedom, and that number is the whole point. A hexapod is statically stable, so unlike a quadruped it can lift legs without falling over, which makes it the ideal platform for actually understanding tripod gaits, body pose control and leg inverse kinematics. Freenove leans into that: the kit self balances, walks over uneven ground, and the same open Python codebase philosophy from their 4WD car applies here, so every gait is source you can read. The rotatable head carries a camera and an ultrasonic sensor, and the software adds face recognition, ranging and app control from a phone or computer. It runs on the Pi 5, 4B, 3B+, 3B and 3A+, with the Zero 2 W possible using extra parts. As with the rest of the Freenove range, the Raspberry Pi and batteries are not included, and eighteen servos is a long build with a lot of calibration before your first clean step.

Everything else in this guide moves things around. The RaspTank Metal picks them up. A tracked chassis handles carpet, cable runs and small obstacles far better than wheels, and the four degree of freedom robotic arm on top turns the whole project into mobile manipulation, which is a much richer software problem: you now have a base frame, an arm frame and a camera frame to reconcile. Adeept pairs it with OpenCV vision for target recognition and colour tracking, plus the usual obstacle avoidance, line tracking, real time video streaming and Wi-Fi control, all in Python. The metal version is the one to buy because it adds Raspberry Pi 5 support alongside the 4B, 3B and 3B+ and replaces the acrylic frame; the older laser cut acrylic RaspTank is cheaper and has far more reviews but is Pi 4 era. Adeept rates the build as intermediate, roughly four to six hours, and the Raspberry Pi is not included.

This is the cheapest sensible way to put a Raspberry Pi on wheels, and for a first project that is exactly the right trade. The three wheeled chassis keeps the mechanics simple, two driven wheels and a caster, so nothing about the drivetrain gets in the way of the software. What you actually get to play with is a pan-tilt camera for live video, and an ultrasonic sensor on a servo that Freenove uses for a radar sweep mode, building a simple polar map of the space in front of the robot. That one feature is a better introduction to sensing than another line follower, because it makes the relationship between servo angle and range reading obvious. Control is from a phone or a computer over Wi-Fi, the code is Python and documented at Freenove’s docs site, and it runs on the Pi 5, 4B, 3B+, 3B and A+. As always with this brand, the Pi and batteries are extra. It has fewer sensors than the 4WD car, so buy it on price, not on features.

Want one kit that does everything?

The SunFounder PiCar-X covers driving, vision, line following, cliff detection and voice on one well documented Python platform, with a Robot HAT that solves power and PWM for you.

Want the most hardware per dollar?

The Freenove 4WD Smart Car gives you four wheel drive, a camera and ultrasonic on a servo head, line and light sensors, and the entire Python source on GitHub.

Heading towards ROS 2?

The Hiwonder TurboPi runs Python, OpenCV and ROS 2 on a metal mecanum chassis, which is the closest thing here to a small research platform.

Want to learn walking robots?

The SunFounder PiDog puts twelve servos, a six axis IMU and a full sensor suite under a clean Python API, so gaits and balance become something you can experiment with.

Prefer six legs to four?

The Freenove Big Hexapod is statically stable, so it forgives a half tuned gait, and its eighteen servos plus open Python make tripod walking a readable exercise.

Want the robot to pick things up?

The Adeept RaspTank Metal is the only kit here with a 4-DOF arm, and its tracks handle carpet and thresholds that stop the wheeled kits.

Just testing the water?

The Freenove Three-Wheeled Smart Car is the cheapest way to get a camera and an ultrasonic radar sweep onto a Pi powered chassis and find out if you enjoy it.

Do not own a Raspberry Pi yet?

Add a Pi 4B or Pi 5, a good SD card and a proper supply to your budget before you compare kits. SunFounder also sells PiCar-X and PiDog bundles with a Pi Zero 2 W if you would rather buy once.

Do Raspberry Pi robot kits include the Raspberry Pi?

Usually not. Every kit in this guide ships on its standard listing without the board, and most also leave out the SD card and the batteries. That is the single biggest reason people feel misled by the price. Some vendors do sell bundles that include a board, typically a Raspberry Pi Zero 2 W with an SD card, and SunFounder offers those for both the PiCar-X and the PiDog. Before you compare two kits on price, add a Pi, a fast microSD card, a decent power supply and cells to whichever one is missing them, then compare again.

Which Raspberry Pi model should I use for a robot?

For anything involving OpenCV, a Pi 4B with 4 GB is the sensible floor and a Pi 5 is noticeably more comfortable, especially if you are running face or object detection at a usable frame rate. A Pi Zero 2 W works well for driving, streaming and simple behaviours and keeps the robot light and cheap, but it will struggle with heavier vision pipelines. Two practical warnings: the Pi 5 uses a different, smaller camera connector, so an older kit’s ribbon cable will not fit, and the Pi 5 draws more current, so check that the kit’s power board can actually feed it while the motors are running.

Raspberry Pi or Arduino for a robot project?

Choose by what the robot has to do. An Arduino gives you a deterministic loop, instant boot, microsecond timing and low power, which is why it still wins for line followers, balancing robots and anything with a tight control loop. A Raspberry Pi gives you Linux, a camera stack, OpenCV, Python libraries, networking, SSH and enough compute for machine learning, at the cost of boot time and timing jitter. The honest test is one question: does your project need a camera or a heavy library? If yes, buy a Pi kit. If no, an Arduino chassis will teach you more for less money. Serious builds often use both, with the Pi doing perception and a microcontroller doing the motor loop.

Why does my Pi robot keep rebooting when the motors run?

Almost always power, not software. Motors and servos pull large current spikes when they start or stall, and if they share a rail with the Pi that spike drops the 5 V line below the point where the Pi resets. The fixes, in order: use a power board that gives the Pi its own regulated supply separate from the motor rail, use healthy cells that can deliver the current rather than tired 18650s, keep the power wiring short and thick, add bulk capacitance across the motor supply, and avoid commanding all servos to move at once on a quadruped or hexapod. If the Pi throttles or shows the low voltage warning under load, treat that as the cause rather than a nuisance message.

Are these kits good for ROS 2, and which one should I pick?

Only one of them targets it directly. The Hiwonder TurboPi ships with a ROS 2 stack alongside Python and OpenCV, which makes it the natural choice if you want to learn nodes, topics and navigation on real hardware rather than in simulation. The other kits are Python first, and while you can absolutely wrap any of them in your own ROS 2 nodes, you will be writing the drivers and message plumbing yourself. That is not wasted effort, and doing it for the Freenove 4WD is a genuinely good way to learn how a ROS 2 hardware interface is built, but be clear that you are choosing a project rather than buying a ready platform.