← Back to ROS 2

05 — Holonomic vs Non-Holonomic vs Underactuated

Why some robots can move directly sideways and others must maneuver

This page exists because many planning and control interviews quietly test one deeper idea:

Is the robot free to move in every local direction, or is its motion constrained by its current heading and actuation?

That distinction explains why some robots can follow a path like a cursor on a screen, while others must swing, reverse, or re-approach to reach the same pose.

1. Mathematical definition

Let the robot configuration be:

$$ q \in \mathbb{R}^n $$

Holonomic

A robot is holonomic if its constraints can be expressed as configuration-only constraints, so the allowed motion directions match the independent coordinates of its configuration space.

In practical robotics language, that means:

The robot can move instantaneously in every independent local direction that its configuration space allows.

For a planar omnidirectional robot, you can model:

$$ q = (x, y, \theta) $$

with directly commanded body motion:

$$ \dot{x} = u_x, \quad \dot{y} = u_y, \quad \dot{\theta} = u_\theta $$

Here the robot can translate in $x$, translate in $y$, and rotate independently.

Non-holonomic

A robot is non-holonomic if it has velocity constraints that cannot be integrated into pure position constraints.

These are often written as:

$$ A(q)\dot{q} = 0 $$

If those constraints cannot be converted into a configuration-only condition like:

$$ h(q) = 0 $$

then they are non-holonomic.

For a unicycle or differential-drive robot:

$$ q = (x, y, \theta) $$
$$ \dot{x} = v\cos\theta $$
$$ \dot{y} = v\sin\theta $$
$$ \dot{\theta} = \omega $$

This implies the sideways-velocity constraint:

$$ -\sin\theta\,\dot{x} + \cos\theta\,\dot{y} = 0 $$

That equation means the robot cannot move sideways instantaneously.

Underactuated

A robot is underactuated when it has fewer independent control inputs than the full number of configuration or state variables that matter for motion.

That is a different question from holonomy.

The common interview-safe summary is:

  • Holonomic / non-holonomic asks: what motion directions are locally feasible?
  • Underactuated asks: how many independent controls do I have relative to the motion I want to control?

A quadrotor is the classic example: it is usually discussed as underactuated, not simply as a holonomic mobile robot.

2. ELI5 intuition

Holonomic robot

Think of a robot that moves like a computer game character on ice-free tiles:

  • forward
  • backward
  • left
  • right
  • rotate

If it needs to shift a little to the left, it just does it.

Non-holonomic robot

Now think of a real car.

It can:

  • go forward
  • go backward
  • turn

But it cannot do this:

“Move 20 cm directly sideways into the next parking spot right now.”

It has to maneuver.

One-line kid version

  • Holonomic: “I can go directly where I want.”
  • Non-holonomic: “I may get there, but I have to steer my way there.”

3. Real-world use cases

Holonomic robots

  • Mecanum AMRs in factories where sideways docking is useful.
  • Omni-wheel service robots in labs, hospitals, or demos.
  • Cartesian gantries where motion is independently commanded along axes.

Non-holonomic robots

  • Differential-drive warehouse AMRs because they are simple and robust.
  • Forklifts because approach angle and reversing matter.
  • Autonomous cars because steering limits define feasible motion.
  • Tuggers and AGVs with trailers because turning constraints dominate maneuvering.

Underactuated systems

  • Quadrotors because thrust and attitude are controlled indirectly rather than commanding arbitrary translational motion directly.
  • Fixed-wing aircraft because they cannot strafe like a holonomic robot.
  • Acrobots / pendulum systems because not every joint is directly actuated.

4. Why this matters in planning

If the robot is approximately holonomic in the plane, a planner can often think more simply:

  • free space on the map
  • path length
  • obstacle avoidance

If the robot is non-holonomic, the planner must also care about:

  • heading
  • curvature
  • turning radius
  • reverse policy

That is why a 2D costmap path is not always a drivable path.

This is exactly the intuition behind:

  • NavFn searching mostly in (x, y)
  • Smac / Hybrid-A* searching in (x, y, \theta)

Related reading: 04-navfn-vs-smac-search-spaces.md

5. Comparison table

Type Core idea Mathematical clue Sideways motion right now? Typical example Planning implication
Holonomic Can move directly in each independent local direction Direct local velocity in all relevant DOFs Usually yes, if that DOF exists Omni-wheel base Geometry-first planning often works
Non-holonomic Has non-integrable velocity constraints A(q)\dot{q} = 0 No Differential drive, car, forklift Heading and curvature matter
Underactuated Has fewer independent controls than motion variables of interest Fewer independent inputs than desired motion authority Not the defining test Quadrotor, fixed-wing aircraft Dynamics and control coupling matter

6. Tricky interview questions

1. Is every wheeled robot non-holonomic?

No. Differential-drive and car-like robots are non-holonomic, but mecanum and omni-wheel robots are often modeled as holonomic in the plane.

2. If a non-holonomic robot can still reach the goal, why call it constrained?

Because the constraint is about instantaneous feasible velocity directions, not necessarily eventual reachability.

3. Why does heading matter so much for non-holonomic robots?

Because heading determines which local velocities are physically feasible.

4. Is underactuated the same as non-holonomic?

No. Underactuation is about missing independent control inputs. Non-holonomy is about motion constraints that forbid arbitrary instantaneous directions.

5. Is a quadrotor holonomic?

In full dynamics, the better answer is that it is underactuated. In simplified planning layers, people may sometimes approximate its position control as if it were free in 3D.

6. Give the cleanest one-sentence distinction

A holonomic robot can move instantaneously in every allowed local direction; a non-holonomic robot must maneuver because some velocity directions are forbidden.

7. Practical takeaway

When someone says a path is valid, ask:

  1. Valid in the map?
  2. Valid for the robot’s heading and turning radius?
  3. Valid under the platform’s actuation limits?

That is the bridge from geometry to real robot motion.

Companion exercise: Exercise08 — Holonomic, Non-Holonomic, and Underactuated Interview Traps