Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields

arXiv:2601.13273 · eess.SY, cs.SY · Submitted 2026-01-19 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields".

Dev: Here's the gist of the paper. Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields provides a complete motion-planning mechanism that ensures target tracking and obstacle avoidance in a cluttered environment.

Rosa: The core contribution is a general treatment of control point selection, enabling a systematic and flexible design of boundary behavior which yields smooth potential surfaces expressed as harmonic functions over an arbitrary polyhedral decomposition of the configuration space.

Dev: Problem Formulation and Approach The paper addresses the challenging task of constructing a potential field surface free of spurious local minima and maxima.

Rosa: This is achieved by decomposing the feasible space into a union of cells and prescribing local harmonic functions within each cell to ensure a feasible path from start point to destination.

Paper summary: Rosa: So, wrapping up, what's the real meaning here beyond just another navigation algorithm? It seems like this paper provides a concrete mechanism for handling cluttered environments by systematically building potentials on decomposed spaces and steering agents along those defined paths.

Dev: I think the key is that they show how to explicitly compute these harmonic potential surfaces from cardinal B-splines, which allows for a direct link between the boundary description and the resulting control policy.

Taro: What this means practically is that we can design environments where we know exactly where safe paths exist and then use this method to ensure our agents follow them through those specific sequences of cells.

Rosa: It’s about taking the abstract idea of a potential field and making it something concrete, something you can actually implement in a real robot or system operating in a complex space.

Dev: The paper is focused on how control point selection influences the boundary behavior, which is crucial because that’s where you control whether the resulting surfaces are smooth or if they have those undesirable local minima or maxima.

Taro: If you’re trying to test autonomy in unpredictable settings, this framework gives you a way to structure the problem so the system doesn't just randomly wander but actively funnels itself toward the target along a pre-vetted chain of cells.

Rosa: So, this work by Nicu and his team is showing that you can use harmonic functions over polyhedral decompositions to create robust motion planning solutions for agents in cluttered spaces.

Dev: It’s a strong demonstration of how careful mathematical construction, using things like Fourier series for the boundary conditions, can lead to a control action with reasonable performance metrics, like that mean norm of thirteen point nine four and standard deviation of eleven point six three we saw earlier <ref:2601.13273#pg2>.

Taro: The limitation I see is that this method relies on you having already constructed that feasible graph based on unique cell facets that don't appear in the obstacle descriptions; if you can't define those initial cells well, the whole path finding part might get stuck.

Rosa: That makes sense, it puts a lot of pressure on the initial setup of the environment model before you even start running the navigation algorithm.

Dev: It’s a very structured approach to motion planning that moves away from just reactive obstacle avoidance toward a more planned, path-constrained movement within known safe regions.

Conclusion: Rosa: So, this paper is about using these spline-based harmonic potential fields to help robots navigate messy spaces safely by making sure they stay on a planned route while avoiding obstacles.

Dev: Right, and what's interesting is how they handle the math behind building those surfaces—they use a specific way to define control points that lets them make the resulting potential surface really smooth.

Taro: It seems like the authors are focusing heavily on controlling where those control points are chosen because that directly impacts whether you get nice, smooth surfaces or these weird little dips and bumps you don't want.

Rosa: Exactly, so they're showing how you can systematically design the boundary behavior to keep things well-behaved across an entire space decomposition.

Dev: And for my part, I’m looking at the control policy they derive; it takes the negative gradient of that potential and mixes it with how fast you are moving compared to the surface itself.

Taro: That mixing term is what allows the agent to follow those prescribed paths, kind of like having a guide telling you exactly where to steer based on your current speed.

Rosa: It’s about creating a path-following mechanism that's robust, even when the environment is cluttered and full of things that could block your way.

Dev: The results they got with the control action—the mean norm being around thirteen point nine four and the standard deviation being eleven point six three—those numbers show it’s performing reasonably well compared to other methods out there.

Taro: But I gotta ask, how does this hold up when you introduce a sudden, unexpected obstacle that wasn't part of the original environment model?

Rosa: That’s a fair question for the real world; the paper focuses on funneling things along a sequence of cells rather than handling completely new surprises.

Dev: So it’s basically a very structured way to enforce safety by pre-calculating safe sequences and forcing the agent to stick to them, even if it means slightly slower travel.

Taro: If you look at the authors, they seem really focused on that systematic design of those control points; I wonder how flexible this system is when you have a totally different type of obstacle geometry.

Theodor-Gabriel Nicu, Florin Stoican, Daniel-Mihail Ioan, Ionela Prodan

National University of Science and Technology Politehnica Bucharest · University Grenoble Alpes

eess.SY, cs.SY

Submitted: 2026-01-19

Updated: 2026-11-03

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 82/100

The gist: Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields provides a complete motion-planning mechanism that ensures target tracking and obstacle avoidance in a cluttered…

Key concepts

Harmonic Potential Surface
This is a smooth landscape created over the configuration space that guides the agent. It is mathematically defined as a harmonic function, meaning it satisfies Laplace's equation, which ensures no local traps (minima or maxima) exist. This surface dictates where the robot should ideally move to avoid obstacles and reach its goal smoothly.
Polyhedral Decomposition
The paper divides the complex configuration space into a collection of simpler shapes called cells. These cells form a polyhedral decomposition, which is a systematic way to map out the environment. By defining local harmonic functions within each cell, the researchers can ensure that paths exist between these defined regions.
Cardinal B-splines
These are special curves used to define the boundaries of the potential surface. They have useful properties like local support and partition of unity, which help in smoothly connecting different parts of the surface. These splines are transformed into Fourier series to help construct the final harmonic function.
Control Policy Derivation
The control action is calculated by combining two parts: one that pushes the agent towards a lower potential (the negative gradient) and another that corrects for any difference between the actual speed and the desired speed on the surface. This combination ensures safe movement along the defined path.

Terminology

Summary

Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields provides a complete motion-planning mechanism that ensures target tracking and obstacle avoidance in a cluttered environment. p. 0

Where the paper says it

IEEE TRANSACTIONS AND JOURNALS TEMPLATE 1 Safe Navigation in Cluttered Environments Via Spline-Based Harmonic Potential Fields Theodor-Gabriel Nicu, Student Member, IEEE, Florin Stoican, Member, IEEE, Daniel-Mihail Ioan, Member, IEEE, and Ionela Prodan, Member, IEEE Abstract— We provide a complete motion-planning mechanism that ensures target tracking and obstacle avoidance in a cluttered environment. For a given polyhedral decomposition of the feasible space, we adopt a novel procedure that constrains the agent to move only through a prescribed sequence of cells via a suitable control policy.

The core contribution is a general treatment of control point selection, enabling a systematic and flexible design of boundary behavior which yields smooth potential surfaces expressed as harmonic functions over an arbitrary polyhedral decomposition of the configuration space. p. 9
Where the paper says it

CONCLUSIONS This paper has detailed the constructive underpinnings of a harmonic function-based motion-planning framework, in which the boundary conditions are induced by cardinal Bspline curves. The core contribution is a general treatment

p. 9
Where the paper says it

CONCLUSIONS This paper has detailed the constructive underpinnings of a harmonic function-based motion-planning framework, in which the boundary conditions are induced by cardinal Bspline curves. The core contribution is a general treatment

Problem Formulation and Approach

The paper addresses the challenging task of constructing a potential field surface free of spurious local minima and maxima. This is achieved by decomposing the feasible space into a union of cells and prescribing local harmonic functions within each cell to ensure a feasible path from start point to destination. p. 1

Where the paper says it

PROBLEM STATEMENT Constructing a potential field surface free of spurious local minima and maxima is a challenging task [14], [15], [11]. In this context, harmonic functions, solutions of the Laplace equation on a prescribed domain, are a popular choice [6], as they address the problem in two stages: i) decomposing the free space into a union of cells and ii) prescribing, via local harmonic functions, the behavior within each cell so as to ensure a feasible path from the start point to the destination.

The method relies on constraining the agent’s trajectory to follow a precomputed sequence of feasible cells, denoted as an agent avoids the obstacles and arrives at the destination.

Harmonic Surface Construction

The potential surface is constructed by taking a harmonic function solution over a polytope P, where P is defined using a diffeomorphic and monotonous mapping φ(q) that pulls the agent onto the unit disk. The boundary condition for this harmonic function is induced by cardinal B-spline curves, which are characterized by properties like local support and partition of unity. p. 1

Where the paper says it

PRELIMINARIES ON CARDINAL B-SPLINES To characterize the potential field within each cell, we consider cardinal B-splines whose properties are detailed next. The cardinal splines Bk,p,Z(t), of degree p ≥ 1, are obtained recursively from relations M0(θ) = ( 1, ∀θ ∈ [0, 1), 0, otherwise, (2a) Mp(θ) = θ p Mp−1(θ) + p + 1 − θ p Mp−1(θ − 1), (2b) Bk,p,Z(θ) = B0,p,Z(θ − k) = Mp(θ − k), ∀k ∈ Z, (2c) with the properties of (given in their ‘local support’ form [19]): P1) Local support: Bk,p,Z(θ) = 0, ∀θ /∈ [k, k + p + 1); (3) P2) Partition of unity: Xm k=m−p Bk,p,Z(θ) = 1, ∀θ ∈ [m, m + 1); (4) P3) The Fourier transform of (2b) is given by F Mp(θ) (ω) = 1 − e −jω jω p+1 .

The resulting harmonic surface is given by the complex expression in equation (44). p. 1
Where the paper says it

In this context, harmonic functions, solutions of the Laplace equation on a prescribed domain, are a popular choice [6], as they address the problem in two stages: i) decomposing the free space into a union of cells and ii) prescribing, via local harmonic functions, the behavior within each cell so as to ensure a feasible path from the start point to the destination. Henceforth, consider a space cluttered with obstacles Oj , R 2 \ [ j Oj = [ ℓ Pℓ, (1) decomposed into a collection of polyhedral, non-overlapping regions Pℓ that form a polyhedral complex [27].

Control Policy Derivation

The control policy for the double integrator dynamics q¨ = u(q) is derived using a combination of the normalized negative gradient of the potential and its deviation between instantaneous velocity and surface gradient. The resulting control action is given by u(q) = K (X(q) − q˙) + X˙ (q), where K > 0 denotes the “velocity regulation”. p. 8

Where the paper says it

[6] adapts this to the double integrator case (48) by, first, denoting the normalized negative gradient of the potential (19) as3 X(q) = − ∇qγP ∥∇qγP ∥ = − ∇⊤ q φ · ∇φ(q)γ ∥∇⊤ q φ · ∇φ(q)γ∥ (49) and, second, its ‘deviation between instantaneous velocity and surface gradient’ as X˙ (q) = (−∇qX) ⊤ q˙. Put together, these give the control action u(q) = K (X(q) − q˙) + X˙ (q), (50) where K > 0 denotes the “velocity regulation”.

Trajectory Computation

The trajectory computation involves several steps outlined in Algorithm 1, starting with constructing a feasible graph G based on unique cell facets that do not appear in obstacle descriptions. A shortest-path algorithm is then applied to link the initial facet Fstart to a final facet Fend to obtain the sequence of cells L. p. 8

Where the paper says it

, the facet pairs belonging to a common cell, E = (Fi , Fj ) : Fi , Fj ∈ N . (52) Applying any shortest-path algorithm to link4 an initial facet Fstart to a final facet Fend, we obtain the sequence {Fk1 , .

Finally, the switched closed-loop dynamics (55) are integrated over time Tend to obtain the feasible trajectory q⋆(t). p. 8
Where the paper says it

Output: feasible trajectory q ⋆ (t), where q ⋆ (t) ∈ ∪ / jOj , ∀0 ≤ t ≤ Tend, with q ⋆ (0) ∈ Fstart and q ⋆ (Tend) ∈ Fend. 1: Construct the feasible graph G, as in (51)–(52); 2: Find shortest path through the graph G as in (53); 3: for i = 1 : M − 1 do 4: select ℓi ∈ L, as in (54); 5: obtain β(q) as in (16) and φ(q) as in (18), for Pℓi ; 6: find θ0, θ1 such that φ(Fki+1 ) maps the corresponding unit circle arc; 7: get k, k as in (42); 8: compute γ(r, θ) as in (44) and γP (q) as in (19); 9: compute control action u(q) as in (50) and apply it to (48) to obtain closed-loop dynamics (55); 10: end for 11: Integrate the switched closed-loop dynamics (55) to obtain the trajectory q ⋆ (t).

Conclusion and Performance

The paper concludes that this framework successfully funnels the agent safely along the chain of cells from the starting point to the target. The illustrative example demonstrates that trajectories remain within the feasible space and safely reach the target. p. 9

Where the paper says it

We observe the desired behavior: the trajectories ( - our method, - method from [7]) remain within the feasible space, i. within the cells composing the path ( ) and safely reach the target.

The method achieves a mean norm of 13. p. 9
Where the paper says it

6 illustrates the control actions (the rhs of (55)), corresponding to the trajectory illustrated in Fig. We observe behavior comparable to [7] in terms of mean and standard deviation for both the norm and the individual components of cell index (ℓi ∈ L, as in (54)) 1 3 9 10 15 16 17 node index (cell facet, as in (53)) 3 1 0 0 0 0 0 0 1 1 1 0 0 0 0 0 6 0 1 1 0 0 0 0 16 0 0 1 1 0 0 0 17 0 0 0 1 1 0 0 25 0 0 0 0 1 1 0 27 0 0 0 0 0 1 1 28 0 0 0 0 0 0 1 TABLE I: Example of path between nodes 3 and 28 −6 −3 0 3 6 −6 −3 0 3 6 −6 −4 −2 x[m] y[m] γ P (q) Fig.

94 and a standard deviation of 11. p. 9
Where the paper says it

6 illustrates the control actions (the rhs of (55)), corresponding to the trajectory illustrated in Fig. We observe behavior comparable to [7] in terms of mean and standard deviation for both the norm and the individual components of cell index (ℓi ∈ L, as in (54)) 1 3 9 10 15 16 17 node index (cell facet, as in (53)) 3 1 0 0 0 0 0 0 1 1 1 0 0 0 0 0 6 0 1 1 0 0 0 0 16 0 0 1 1 0 0 0 17 0 0 0 1 1 0 0 25 0 0 0 0 1 1 0 27 0 0 0 0 0 1 1 28 0 0 0 0 0 0 1 TABLE I: Example of path between nodes 3 and 28 −6 −3 0 3 6 −6 −3 0 3 6 −6 −4 −2 x[m] y[m] γ P (q) Fig.

63 for the control action, which is comparable to existing methods. p. 9
Where the paper says it

6 illustrates the control actions (the rhs of (55)), corresponding to the trajectory illustrated in Fig. We observe behavior comparable to [7] in terms of mean and standard deviation for both the norm and the individual components of cell index (ℓi ∈ L, as in (54)) 1 3 9 10 15 16 17 node index (cell facet, as in (53)) 3 1 0 0 0 0 0 0 1 1 1 0 0 0 0 0 6 0 1 1 0 0 0 0 16 0 0 1 1 0 0 0 17 0 0 0 1 1 0 0 25 0 0 0 0 1 1 0 27 0 0 0 0 0 1 1 28 0 0 0 0 0 0 1 TABLE I: Example of path between nodes 3 and 28 −6 −3 0 3 6 −6 −3 0 3 6 −6 −4 −2 x[m] y[m] γ P (q) Fig.

The code implementing this method is available on gitlab. p. 9
Where the paper says it

18 TABLE II: Input statistics for the example shown in Fig. The code is available at gitlab.

com/replan/replan-public. p. 9
Where the paper says it

The code is available at gitlab. com/replan/ replan-public.

git.

Pruning redundant reasoning tokens

The paper provides a complete motion-planning mechanism that ensures target tracking and obstacle avoidance in a cluttered environment. The core contribution is a general treatment of control point selection, enabling a systematic and flexible design of boundary behavior which yields smooth potential surfaces expressed as harmonic functions over an arbitrary polyhedral decomposition of the configuration space.

How the training signal is built

The harmonic function h(θ) is represented as a Fourier trigonometric series derived from the Fourier transform of cardinal B-splines. The coefficients A0, An, and Bn are computed using relations derived from properties of the spline functions and Fourier analysis. p. 6

Where the paper says it

Replacing p with p − 1 in (23) and using that the spline functions Mp−1(·) are positive, we obtain the sign table: 0 λk λ(k + p) λk λ(k + p) 2π 0 0 − − 0 0 + + 0 0 . θ sgn( dh(θ) dθ ) The underlying assumption in constructing this table is that the order relations 0 < λk < λ(k +p) < λk < λ(k +p) < 2π hold.

These coefficients are then used to express the harmonic potential surface γ(r, θ) in terms of polylogarithm functions. p. 7
Where the paper says it

For better illustration, the cardinal splines are vertically shifted by −1; those multiplied by Pk = −1 are shown in blue, while those multiplied by Pk = 0 are shown in light gray. 3b we plot the harmonic potential surface (44), where the polylogarithm functions are approximated by truncating the series (33) after the first 15 terms.

Where the gains come from

The construction of the control action u(q) involves two components: a term proportional to the normalized negative gradient and a term capturing the deviation between instantaneous velocity and surface gradient. The method ensures that trajectories are constrained to exit through the prescribed facet, as illustrated by the waterfall effect along a path segment.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

1

Improve motion planning by constraining agent movement to a prescribed sequence of cells via a suitable control policy, ensuring safe navigation in cluttered environments through funneling along the chain of cells from start to target.

2

Implement obstacle avoidance and target tracking by constructing harmonic potential surfaces where the minima and maxima are confined to the boundary of regions, which is achieved by prescribing Dirichlet boundary conditions given as a cardinal B-spline curve.

3

Develop a control policy that drives the system along the negative gradient of these harmonic surfaces, specifically utilizing the switched dynamics q'' = K (X(q) − q') + X'(q) to track integral curves of normalized gradients.

4

Enhance trajectory computation by constructing a feasible graph G where nodes are unique cell facets and edges represent facet pairs belonging to a common cell, then applying any shortest-path algorithm to determine the sequence of cells through which the agent must pass.

Abstract

We provide a motion-planning framework for target reaching and obstacle avoidance in cluttered environments. Given a polyhedral decomposition of the feasible space, the agent is guided through a prescribed sequence of cells by a suitable control policy. For each cell, we construct a potential induced by a harmonic function on the unit disk whose Dirichlet boundary condition is represented by a cardinal B-spline curve. A detailed analysis of the curve periodicity, support, and control-point selection provides an explicit analytical construction of these potentials and the associated vector fields. We further characterize inter-cell control continuity and forward invariance of non-exit facets, and introduce a smoothed control law ensuring continuous switching between independently constructed neighboring cells. Numerical results illustrate that the resulting framework safely guides the agent along the prescribed cell sequence from the initial point to the target.

Sources

Related papers