Smooth Move: Taming Trajectories with Polynomials

Smooth Move: Taming Trajectories with Polynomials · ... and another thing ...

... and another thing ...

2026-08-03

robots

3dprint

python

maths

This article is all about movement. Of a vehicle, or a robot, or something like a 3D printer.

A 3D printer is really just a specialized robot with a hot glue gun which it can move in multiple directions while squirting out molten plastic.

Position, Velocity, Acceleration …

Our robot, or print head, has a position in space and also a velocity which is how fast it is moving and an acceleration which is how rapidly that velocity is changing.

Velocity is just the derivative of position: how fast is our position changing. Acceleration is the derivative of velocity: how rapidly our velocity is increasing or decreasing.

These are all vectors in that they have both magnitude and direction but for (many) 3D printers and machines each axis is a separate mechanism, so this article is for now going to talk about them as if they were scalars, eg: just a positive or negative number.

… Jerk?

Jerk is the derivative of acceleration, which might seem like a pretty abstract thing to be worried about.

But imagine a mass in a box under constant acceleration. Springs inside the box are pushing the mass to cause it to accelerate too. But if we change the size or direction of the acceleration, the mass is going to slide around until it reaches a new equilibrium.

Now imagine the box is your skull and the mass is your brain. Jerk is real!

Higher Orders

There’s also higher derivatives which are sometimes called Snap, Crackle and Pop.

Snap (sometimes called Jounce) is the rate of change of Jerk; Crackle is the rate of change of Snap, etc. I don’t have as neat an illustration of what these physically mean but there seems to be a consensus that they, and presumably even higher derivatives, have an effect on vibration and so on of mechanisms, and so they should be minimized too.

Discontinuities

Our 3D printer or similar machine follows a path made up of many segments. Some segments have fixed position and velocity: while the printer is extruding or the CNC mill is cutting, it goes in a set direction at a set speed.

Other parts of the movement are more free: the printer just has to get to the right place for the start of its next fixed segment, as rapidly as is practical1.

The problem is smoothly transitioning between segments. A discontinuity in velocity requires a large acceleration. A discontinuity in acceleration requires a large jerk. And so on. When we combine segments into a path, it is important to match the ends to prevent these discontinuities. So our apparently “free” transport segments are actually critical to support smooth transitions in and out of our “working” segments.

Transitions & Trajectories

So let’s look at a way to make our transitions smooth.

What we’re looking for is some kind of Sigmoid Function which has the right general shape.

The most obvious sigmoid functions is the Logistic Function which does indeed transition smoothly between 0 and 1 and has well defined derivatives, but unfortunately it only converges towards 0 and 1 whereas we want to get our transition over and done with in finite time.

There are other suitable functions though, including polynomials.

SmoothStep …

SmoothStep is a family of polynomial functions which smoothly transition across the range [0,1] in the domain [0,1] in a Sigmoid shape.

$ S_1(t) = \begin{cases}0, & if\ t \leq 0 \\ 3t^2 - 2t^3, & if\ 0 \leq t \leq 1 \\ 1, & if\ 1 \leq t\end{cases} $

Smoothstep is a polynomial function and so the first derivative $ S^\prime_1 $, is polynomial too:

$ S'_1(t) = \begin{cases}0, & if\ t \leq 0 \\ -6t^2 + 6t, & if\ 0 \leq t \leq 1 \\ 0, & if\ 1 \leq t\end{cases} $

and has the handy property that it is neatly zero at both ends. However, $ S''_1 $ does not have this property. Our velocity at each end of our movement is zero, but our acceleration is not.

… and SmootherStep …

If we want acceleration to be zero at the beginning and end of our trajectory, we can use Smootherstep, which is a fifth-order polynomial with this property:

$ S_2(t) = \begin{cases}0, & if\ t \leq 0 \\ 6t^5 - 15t^4 +10t^3, & 0 \leq t \leq 1 \\ 1, & 1 \leq t\end{cases} $

$ S'_2(t) = \begin{cases}0, & if\ t \leq 0 \\ 30t^4 - 60t^3 + 30t^2, & 0 \leq t \leq 1 \\ 0, & 1 \leq t\end{cases} $

$ S''_2(t) = \begin{cases}0, & if\ t \leq 0 \\ 120t^3 - 180t^2 + 60t, & 0 \leq t \leq 1 \\ 0, & 1 \leq t\end{cases} $

… and SmoothnStep

The SmoothStep function can be worked out to an arbitrary depth, for example $ S_6 $ is a 13th-order polynomial:

$ S_6(t) = \begin{cases}0, & t \leq 0 \\ 924t^{13} - 6006t^{12} + 16380t^{11} - 24024t^{10} + 20020t^9 - 9009t^8 + 1716t^7, & 0 \leq t \leq 1 \\ 1, & 1 \leq t\end{cases} $

For the $ n $-th Smoothstep function, all derivatives up to the $ n $-th derivative start and end at zero:

$ S^{(m)}_n(0) = S^{(m)}_n(1) = 0...

Smooth Move: Taming Trajectories with Polynomials

Related Articles

(no title)

Scientists reverse brain aging, with a nasal spray

NHS apologises and admits Palantir have access to identifiable patient data

AI has torched the market for junior programmers

Al Vigier: Canada's AI strategy shouldn't include secret Palantir bills