Blog · 2026-08-15 · Kinematics
The drivetrain decision shapes everything downstream — your motor count, your odometry accuracy, your software complexity and whether the robot can get out of a corner. Here's an honest comparison of the five options actually worth considering.
Drivetrain choice tends to get made early, casually, and permanently. Someone sees a mecanum robot gliding sideways on video and orders four mecanum wheels; six weeks later they're fighting slip-ruined odometry and wondering why the robot can't drive straight. The mechanisms genuinely differ in what they're good at, and the differences are predictable. This is the comparison, with the trade-offs stated plainly.
| Drivetrain | Motors | Spins in place | Moves sideways | Odometry quality | Best for |
|---|---|---|---|---|---|
| Differential (2WD + caster) | 2 | Yes | No | Excellent | Indoor robots, navigation, learning |
| Skid-steer (4WD) | 4 (or 2 geared) | Yes, with scrub | No | Poor | Rough ground, traction, payload |
| Mecanum / omni | 4 | Yes | Yes | Poor | Flat floors, tight spaces, competition |
| Tracked | 2 | Yes, heavy scrub | No | Very poor | Obstacles, loose terrain, stairs |
| Ackermann (car-like) | 1–2 + steering | No | No | Good | Outdoor, high speed, scale vehicles |
Two powered wheels on a common axis plus one or two casters. It is the cheapest to build, the simplest to control, and — the point people miss — it has the best odometry of any drivetrain, because both powered wheels roll cleanly in the direction they point. No scrub, no slip beyond normal traction limits, so encoder counts genuinely correspond to ground distance. Every navigation stack, SLAM package and tutorial assumes this model, and the maths is two equations wide, as laid out in Differential Drive Kinematics.
Its weaknesses are real but narrow. Casters wander on carpet transitions and can jam on cables. Weight distribution matters: too much on the caster and the drive wheels lose traction; too little and the robot rocks. And it cannot move sideways, so tight alignment tasks require a manoeuvre. For an indoor robot that needs to navigate reliably, none of these outweigh the odometry advantage.
Four powered wheels, no steering — turning happens by driving the sides at different speeds and letting the wheels drag sideways. The gains are genuine: double the traction, no caster to snag, better obstacle handling, and a stable rectangular footprint that carries payload well. It's why nearly every rough-terrain and agricultural robot uses it.
The price is paid in scrub. Because every turn drags four contact patches sideways, the actual turn rate falls well short of what the kinematics predict — commonly 20–30% short on grippy surfaces, and the error changes with the surface, which is what makes it so hard to calibrate away. Skid-steer robots also consume noticeably more current while turning, and chew tyres. If you build one, budget for a gyro or IMU from the start: fusing gyro heading with wheel distance is not a luxury here, it's the only way to get usable heading.
Mecanum wheels have angled rollers around their circumference; drive all four at the right relative speeds and the diagonal force components combine into sideways or diagonal motion. Omni wheels do something similar with perpendicular rollers, usually in a three- or four-wheel arrangement. Holonomic motion — moving in any direction without rotating first — is genuinely useful for tight spaces and alignment tasks, and it looks superb.
The conditions attached are steep. Mecanum wheels need flat, clean, hard floors; the small rollers fall into gaps, catch on carpet and stop dead on debris. All four wheels must stay loaded, so a suspension or compliant mount is close to mandatory on anything but a perfect surface. Efficiency is lower, since a fraction of every wheel's force is spent fighting the other wheels. And odometry is poor by construction: the rollers slip constantly, which is the mechanism working as intended but means wheel counts are a weak estimate of motion. Competition robots on smooth arena floors are the ideal case — hence their popularity in FIRST and VEX — but for a general indoor robot they solve a problem most builds don't have.
Tracks spread weight over a large contact area and climb things wheels can't — kerbs, rubble, stairs with the right geometry. For search-and-rescue-style robots and loose terrain they're unmatched. But every turn is pure scrub along the entire track length, so heading estimation from track speed is nearly worthless, turning consumes serious current, and tracks throw themselves off sprockets at inconvenient moments unless tensioning and alignment are done well. Choose tracks when the terrain demands them, not for the aesthetic.
Car-like steering — powered rear wheels, steered front wheels — is efficient, stable at speed and kind to tyres, since all four wheels roll in the direction they point. The disqualifying limitation for most hobby robots is the minimum turning radius: an Ackermann robot cannot spin in place or reverse out of a dead end without a three-point turn, and path planning becomes substantially harder as a result. It earns its place on larger outdoor platforms and scale vehicles where speed and efficiency matter more than manoeuvrability.
Work backwards from constraints rather than forwards from appeal. Where does it drive? Smooth indoor floor opens every option; carpet rules out mecanum; rough ground points to skid-steer or tracks. Does it need accurate autonomous navigation? If yes, differential drive's odometry advantage is worth more than any other feature on the list. Does it need to translate sideways? Only mecanum and omni do, and only on good floors. How many motor channels can you power and afford? Four-motor drivetrains double your driver cost and current draw — worth checking against the Power Budget Calculator before committing. How much software complexity can you carry? Differential is two equations; mecanum is a matrix plus a slip problem.
For most people building an indoor autonomous robot, the honest answer is: build a differential drive, get the navigation working, and only move to something more exotic when a specific limitation actually blocks you. The simplest drivetrain that meets the requirement leaves the most budget — of money, current and attention — for everything else.
The drivetrain decides the geometry; the motors still have to move it. Once you've picked a configuration, the torque requirement comes from the Motor Sizing Calculator, the wheel speeds from the Differential Drive Calculator, and the tip-over margin — which changes a lot between a narrow two-wheeler and a square four-wheeler — from the Robot Stability Calculator.
Yes — power two and leave two as unpowered idlers or casters, giving differential-drive behaviour with a four-corner footprint. It's a good compromise for stability without the scrub penalty, provided the unpowered wheels genuinely swivel or the robot will fight itself in turns.
Three-wheel omni is mechanically simpler, needs no suspension to keep contact, and gives full holonomic motion — but the triangular footprint is less stable and payload mounting is awkward. On a flat floor it's an underrated option.
Kinematically it's a differential drive, with an added inverted-pendulum control problem stacked on top. Wonderful for learning control theory, poor as a platform for carrying anything or navigating reliably.
Drivetrain decisions are expensive to reverse — they determine the chassis, the motor count and often the whole electrical layout — so it's worth spending a weekend de-risking one. Two cheap tests catch most regret. The surface test: take a single wheel of the type you're considering, load it with roughly the weight it will carry, and drag it across every surface the robot will meet — the carpet, the threshold strips, the gap between floorboards, the cable you keep meaning to tidy. Mecanum rollers and small casters fail this test visibly and immediately, and it costs nothing to discover it before the chassis exists.
The turning test: mock up the footprint with the intended track width and wheelbase, weight it realistically, and push it through the tightest turn the robot will actually need. Skid-steer platforms feel dramatically different at a square footprint versus a long narrow one, and a cardboard mock-up conveys that in seconds where a spreadsheet never will. Both tests take an hour and routinely change people's minds — which is exactly what you want a test to do while the design is still on paper.
Match the drivetrain to the floor and the mission, not to the video that inspired the build. Then check the numbers: wheel speeds, stability margin, and the kinematics that tie them together.