Continuous Spline Parameterization for Kinematic Trajectory Smoothing

341 concepts272 papers3 proposals

Imagine an autonomous quadrotor navigating a dense forest or a self-driving car turning a sharp corner. To find a path from start to finish, standard robotic planners—such as $A^*$ or Rapidly-exploring Random Trees ($RRT^*$)—discretize the environment into a grid or a network of geometric nodes. They output a sequence of waypoints connected by straight lines.

Geometrically, these paths are perfectly valid. Physically, they are impossible to execute.

A piecewise-linear path contains sharp vertices. For a physical vehicle to instantly change its direction at a vertex, it would require infinite acceleration and instantaneous torque. In the real world, attempting to follow these jagged paths causes severe mechanical stress, violent vibrations, passenger discomfort, or traction loss as wheels slip trying to execute discontinuous changes in velocity. As Choi et al. observe, linear connections between discrete points compromise vehicle stability, making continuous curvature transitions a strict requirement for realistic physical dynamics.

To bridge the gap between discrete geometric planning and continuous physical execution, modern robotics relies on continuous spline parameterization. This approach transforms jagged, discrete polylines into smooth, mathematically differentiable trajectories that inherently respect the kinematic and dynamic limits of physical hardware.


The Geometry of Smoothness: Understanding Continuity

To understand why spline parameterization works, we must look at the mathematical definition of smoothness, known as differentiability class ($C^n$).

A standard path generated by a grid planner is only $C^0$ continuous: the path is unbroken, but its first derivative (velocity) is discontinuous at every corner. If a robot attempts to follow a $C^0$ path at a constant speed, its acceleration spikes to infinity at the turning points.

Jagged Path (C0 Continuous):   ───┐
                                  └───> (Infinite acceleration at corners)

Smoothed Path (C2 Continuous):   ───╮
                                    ╰───> (Continuous velocity and acceleration)

By applying higher-order parametric curves—such as Bézier curves or B-splines—to the discrete vertices, planners can elevate the trajectory to $C^1$ (continuous velocity) or $C^2$ (continuous acceleration) continuity.

  • Bézier Curves: Dai and Yue et al. demonstrate that fitting third-order Bézier curves to the turning points of a raw path eliminates sharp transitions. This ensures continuous curvature, allowing mobile robots to maintain a more constant velocity and avoid the need to stop and pivot at every intersection.
  • B-Splines: For more complex trajectories, cubic or quintic B-splines are preferred. Liu et al. leverage cubic B-splines to guarantee that generated paths respect physical constraints like minimum turning radii and smooth velocity transitions. Because these curves are twice-differentiable ($C^2$), the physical actuators experience smooth, continuous changes in force rather than jerky impulses.

The Optimization Trick: Why B-Splines Are Fast

If smoothing a path required solving a massive, constrained optimization problem across the entire trajectory every time an obstacle moved, robots would freeze up from computational overload. The widespread adoption of B-splines in real-time applications comes down to two unique mathematical properties: local support and non-interpolation.

Traditional polynomial interpolation forces a curve to pass exactly through every single waypoint. If you change one waypoint to avoid an obstacle, the entire curve shifts, requiring a full recalculation of the trajectory.

B-splines, however, are approximating splines. The curve is guided by a set of control points but does not strictly pass through them. Instead, the curve is contained within the convex hull of its control points.

Control Point (Unconstrained)
       ○  . . . . . . . 
      . \               .
     .   \               .
    .     \               .
   ○       \               ○ Control Point
  Control   \  Actual     /
   Point     ╰───────────╯ 
             Smooth Curve

This design provides two massive advantages for real-time robotics:

  1. Unconstrained Optimization: As Usenko et al. explain, because B-splines do not need to strictly intersect their control points, their mathematical formulation inherently guarantees boundary continuity. This allows planners to use fast, unconstrained optimization algorithms, bypassing the computationally expensive boundary constraint enforcement required by traditional polynomials.
  2. Local Support: Moving a single control point only alters the trajectory locally. If a drone detects a sudden obstacle, it only needs to adjust the nearby control points. The rest of the global trajectory remains mathematically untouched and stable.

This computational efficiency allows frameworks like STORM, developed by Bhardwaj et al., to embed B-spline parameterization directly inside stochastic sampling-based controllers, achieving high-frequency (125Hz) joint-space control that eliminates erratic movements natively. Similarly, Le et al. demonstrate that integrating spline interpolation directly into the initial search phase of a planner removes the need for a separate, expensive post-processing smoothing step altogether.


Beyond Robotics: A Universal Mathematical Tool

The utility of transforming noisy, discrete sequences into continuous, differentiable curves extends far beyond mobile robots and drones. The same underlying mathematics solves problems across diverse engineering domains:

  • Energy Efficiency in Maritime Shipping: Li et al. utilize a two-stage smoothing process (vertex reduction followed by cubic B-splines) to plan ship routes through dynamic ocean currents. Ensuring continuity in speed and acceleration minimizes unnecessary course corrections, directly translating to reduced fuel consumption.
  • Manufacturing and CNC Machining: In five-axis machining, toolpaths must be incredibly smooth to prevent tool chatter and surface defects. Gao et al. introduce double B-spline curve-fitting to synchronize tool movements, reducing machine vibration and ensuring the axes operate safely within their kinematic limits.
  • Medical Data and Signal Processing: Raw data collected from sensors or human demonstrations is notoriously noisy. In functional data analysis, Cai et al. use B-spline smoothing as a pre-processing step to reconstruct continuous underlying signals from noisy, discrete observations. In robotic surgery, the authors of the Endo[PA]R system apply cubic spline interpolation to human surgical demonstrations to filter out physiological tremors, creating smooth, repeatable paths for autonomous execution.

Whether the goal is preventing a drone from crashing due to sudden aerodynamic shifts, saving fuel on an ocean liner, or removing the robotic jitter from an AI-generated sketch (SketchAgent), continuous spline parameterization serves as the essential mathematical translator between the jagged, discrete logic of algorithms and the smooth, continuous physics of the real world.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 250 papers
  2. Version 1 · 2026-10-04 · 190 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_trajectory_smoothing_7408fa0a,
  title = {trajectory smoothing},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/trajectory-smoothing/rdr_7408fa0af6d347f5907fe2916fd1465a},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_trajectory_smoothing_7408fa0a_v2,
  title = {trajectory smoothing},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/trajectory-smoothing/rdr_7408fa0af6d347f5907fe2916fd1465a/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake