Kinematically Feasible Planning for Non-Holonomic Vehicles

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2355 papers428 conceptsfrozen 2026-10-04

Most classical pathfinding algorithms, from A* to basic RRTs, treat a robot as a "holonomic" point—a dimensionless entity that can change direction instantaneously or move sideways as easily as it moves forward. In the real world, however, a car cannot slide laterally into a parking spot, and a fixed-wing drone cannot make a 90-degree turn at a single vertex.

This research direction focuses on bridging the gap between geometric optimality (the shortest path on a map) and kinematic feasibility (the path a physical vehicle can actually execute).

The Point-Mass Fallacy

The foundational challenge, as identified as early as 1977 by Alan M. Thompson in The Navigation System of the JPL Robot, is that algorithms treating robots as points often produce routes that are physically impossible to follow. If a planner ignores a vehicle's width or its minimum turning radius, the resulting "optimal" path might clip a corner or require a maneuver the hardware simply cannot perform.

Charles E. Thorpe noted in Path Relaxation: Path Planning for a Mobile Robot that most grid-based searches lack the constraints necessary to account for steering geometry. This leads to a "failure mode" where a planned route is mathematically valid but mechanically useless.

Understanding Non-Holonomic Constraints

The core of this problem lies in non-holonomic constraints. A system is non-holonomic when its controllable degrees of freedom are fewer than its total configuration parameters. For a car, you have three parameters (x, y, and orientation $\theta$), but you can only directly control two (velocity and steering angle). You cannot change your $y$ position without also changing $x$ and $\theta$.

Jean-Paul Laumond formalized this in Finding Collision-Free Smooth Trajectories for a Non-Holonomic Mobile Robot, showing that for car-like robots, the search space must be transformed to respect these kinematic limits. This often involves using specific geometric primitives:

  • Dubins Curves: The shortest path between two points for a vehicle that only moves forward with a minimum turning radius.
  • Reeds-Shepp Curves: An extension for vehicles that can also move in reverse.

As Ketan Savla and Emilio Frazzoli demonstrate in their work on Dubins Traveling Salesperson Problems, these constraints fundamentally change the complexity of planning. In a holonomic world, the distance between points is a simple straight line; in a Dubins world, the distance depends entirely on the vehicle's starting and ending orientation.

Strategies for Feasible Planning

Researchers have developed several distinct strategies to integrate these physical realities into planning:

1. Decoupled Planning

A common approach is to first find a "rough draft" holonomic path and then refine it. Jean-Claude Latombe pioneered this in A Fast Path Planner for a Car-Like Indoor Mobile Robot, showing that decoupling the problem can make planning 10 to 100 times faster than searching the full high-dimensional state space from scratch.

2. Kinematic-Aware Search and Smoothing

Instead of post-processing a path, algorithms like KSA*, proposed by Sangyol Yoon and Unghui Lee, incorporate kinematic heuristics directly into the search. Similarly, Ziyue Feng and Yu Chen use "imaginative" simulations of feasible maneuvers to build target trees that ensure every branch of a search is actually drivable.

3. Hybrid Optimization

Modern approaches often combine geometric search with control theory. Junxiao Xue and colleagues utilize Model Predictive Control (MPC) to ensure that the trajectories generated are not just smooth, but also respect the dynamic limits of the vehicle's actuators.

Beyond Single Robots: Swarms and Learning

The challenge intensifies in multi-agent systems. Victoria Tuck and S. Shankar Sastry point out in DEC-LOS-RRT that decentralized models often rely on "single-integrator" kinematics—assuming agents can stop or turn instantly. This is a dangerous simplification for fixed-wing drones or high-speed ground vehicles.

To address this, researchers like Arbaaz Khan are developing differentiable planners that allow Reinforcement Learning models to "feel" the physical constraints of the robot during training. By projecting actions onto a space of feasible velocities, these systems learn to navigate without ever proposing a maneuver that violates the laws of physics.

Why It Matters

Ignoring kinematics isn't just a matter of inefficiency; it is a matter of safety. As Chris Denniston and Stephanie Kemna argue in their work on hazardous environments, a planner that doesn't understand a vehicle's turning circle might plan a path that forces the vehicle into a high-risk zone (like a minefield or a cliff edge) simply because it couldn't make a turn tight enough to stay in the "safe" corridor. By embedding the vehicle's physical DNA into the planning process, we move from abstract geometry to reliable, real-world autonomy.

Papers in this version

Showing 60 of 355 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 253 papers
  2. Version 1 · 2026-10-04 · 355 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_non_holonomic_motion_constraints_d7ae0dc5,
  title = {non-holonomic motion constraints},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-holonomic-motion-constraints/rdr_d7ae0dc57d9947918c330f02f0eb2a25},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_kinematically_feasible_planning_for_non__d7ae0dc5_v1,
  title = {Kinematically Feasible Planning for Non-Holonomic Vehicles},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/kinematically-feasible-planning-for-non-holonomic-vehicles/rdr_d7ae0dc57d9947918c330f02f0eb2a25/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}