2025-11-24T04:16:18.668686

Multi-UAV Swarm Obstacle Avoidance Based on Potential Field Optimization

Hu, Wu, Chen
In multi UAV scenarios,the traditional Artificial Potential Field (APF) method often leads to redundant flight paths and frequent abrupt heading changes due to unreasonable obstacle avoidance path planning,and is highly prone to inter UAV collisions during the obstacle avoidance process.To address these issues,this study proposes a novel hybrid algorithm that combines the improved Multi-Robot Formation Obstacle Avoidance (MRF IAPF) algorithm with an enhanced APF optimized for single UAV path planning.Its core ideas are as follows:first,integrating three types of interaction forces from MRF IAPF obstacle repulsion force,inter UAV interaction force,and target attraction force;second,incorporating a refined single UAV path optimization mechanism,including collision risk assessment and an auxiliary sub goal strategy.When a UAV faces a high collision threat,temporary waypoints are generated to guide obstacle avoidance,ensuring eventual precise arrival at the actual target.Simulation results demonstrate that compared with traditional APF based formation algorithms,the proposed algorithm achieves significant improvements in path length optimization and heading stability,can effectively avoid obstacles and quickly restore the formation configuration,thus verifying its applicability and effectiveness in static environments with unknown obstacles.
academic

Multi-UAV Swarm Obstacle Avoidance Based on Potential Field Optimization

Basic Information

  • Paper ID: 2511.16911
  • Title: Multi-UAV Swarm Obstacle Avoidance Based on Potential Field Optimization
  • Authors: Yendo Hu, Yiliang Wu, Weican Chen (Jimei University, China)
  • Classification: cs.RO (Robotics), cs.MA (Multi-Agent Systems)
  • Publication Date: 2024
  • Paper Link: https://arxiv.org/abs/2511.16911

Abstract

In multi-UAV scenarios, traditional Artificial Potential Field (APF) methods often suffer from irrational obstacle avoidance path planning, resulting in redundant flight paths and frequent heading mutations, with high collision risks among UAVs during avoidance. To address these issues, this study proposes a novel hybrid algorithm combining an improved multi-robot formation obstacle avoidance (MRF-IAPF) algorithm with enhanced APF for single-UAV path planning optimization. The core concepts include: first, integrating three types of interactive forces in MRF-IAPF—obstacle repulsion, inter-UAV interaction, and target attraction; second, introducing refined single-UAV path optimization mechanisms, including collision risk assessment and auxiliary sub-target strategies. When UAVs face high collision threats, temporary waypoints are generated to guide avoidance, ensuring final precise arrival at the actual target. Simulation results demonstrate that compared to traditional APF-based formation algorithms, this algorithm achieves significant improvements in path length optimization and heading stability, effectively avoids obstacles and rapidly recovers formation configuration, validating its applicability and effectiveness in unknown static obstacle environments.

Research Background and Motivation

1. Problems to be Addressed

This research targets three core problems in multi-UAV formation obstacle avoidance:

  • Path Redundancy: Traditional APF methods produce unnecessary detour paths, reducing flight efficiency
  • Heading Mutations: Frequent sharp turns during avoidance affect flight stability and practical executability
  • Inter-UAV Collisions: Trajectory distortion and collision risks caused by inter-UAV interaction forces in formation flight

2. Problem Significance

  • Application Demand: UAV applications in military reconnaissance, emergency rescue, and environmental protection are increasingly widespread, with path planning being a core critical technology
  • Safety Requirements: Multi-UAV cooperative operations must ensure formation stability and avoidance safety
  • Real-time Requirements: Practical application scenarios demand algorithms with low computational cost and rapid response capability

3. Limitations of Existing Methods

While traditional APF methods offer advantages such as low computational cost, fast real-time response, and simple control logic, they suffer from:

  • Discontinuous repulsive potential field functions at influence range boundaries, causing gradient mutations
  • Lack of dynamic assessment of relative positions and collision risks among UAVs
  • Inflexible auxiliary target point settings, insufficient consideration of obstacle relative position information
  • Weakening target attraction force as distance decreases, affecting precise arrival capability

4. Research Motivation

Based on in-depth analysis of existing methods, this paper aims to propose a comprehensive improvement solution by integrating multi-robot formation control and single-machine optimization mechanisms, simultaneously maintaining APF method advantages while systematically addressing its inherent defects in multi-UAV scenarios.

Core Contributions

  1. Multi-UAV Interaction Mechanism: Introduces an inter-UAV interaction force model based on Euclidean distance, interaction velocity, and k-nearest neighbor search mechanism, enabling collision avoidance among UAVs and rapid formation establishment and stable maintenance
  2. Hybrid APF Algorithm: Proposes an improved hybrid artificial potential field algorithm, optimizing the repulsive potential field function to ensure first-order gradient continuity, integrating collision risk assessment mechanism and auxiliary sub-target strategy
  3. Collision Risk Assessment: Implements a collision risk calculation mechanism based on safe distance angle, dynamically evaluating collision threats between UAV flight trajectories and obstacles
  4. Auxiliary Sub-target Strategy: Dynamically generates temporary waypoints based on obstacle relative positions, effectively shortening flight path length and reducing heading angle mutation frequency
  5. Performance Verification: Validates algorithm effectiveness in unknown static obstacle environments using AirSim simulation, achieving significant improvements compared to traditional methods in path length and heading stability (84% reduction in heading angle change frequency)

Method Details

Task Definition

Input:

  • Initial position set of multi-UAV system {P1,P2,...,Pn}\{P_1, P_2, ..., P_n\}
  • Target position set
  • Unknown static obstacle environment
  • Desired formation configuration (ideal distance dd, allowable deviation φ\varphi)

Output:

  • Real-time obstacle avoidance path for each UAV
  • Trajectory maintaining formation configuration

Constraints:

  • Inter-UAV distance constraint: d−φ<∣∣Pj−Pi∣∣<d+φd - \varphi < ||P_j - P_i|| < d + \varphi
  • Communication radius constraint: ∣∣Pj−Pi∣∣<r||P_j - P_i|| < r
  • Turning performance constraint: Avoid large angle mutations
  • Collision avoidance: Maintain safe distance from obstacles and other UAVs

Model Architecture

1. Formation Maintenance Model

Models multi-UAV system as undirected graph G(q,w)G(q,w) based on graph theory:

  • Node set q={q1,q2...qn}q = \{q_1, q_2...q_n\} represents each UAV
  • Edge set w⊆q×qw \subseteq q × q represents inter-UAV connections
  • Adjacency matrix A=[aij]∈Rn×nA = [a_{ij}] \in R^{n×n} describes adjacency relationships

Neighbor Definition: Ni={j∈q:∣∣Pj−Pi∣∣<r}N_i = \{j \in q: ||P_j - P_i|| < r\}

Inter-UAV Interaction Force (Equation 5):

-\beta (\frac{1}{dist}-\frac{1}{d}) (\frac{1}{dist})^2 \overrightarrow{dist}, & dist < d \\ 0, & dist = d \\ \alpha (\frac{2}{1 + e^{-dist}} - 1) (dist - d)\overrightarrow{dist}, & dist > d \end{cases}$$ Where: - When $dist < d$, repulsive force prevents excessive proximity - When $dist = d$, interaction force equals zero, achieving equilibrium - When $dist > d$, attractive force maintains formation - $\alpha, \beta$ are adjustment coefficients #### 2. Obstacle Detection and Localization Employs lightweight traditional image processing methods: 1. Acquire depth and segmentation images 2. Calculate 3D coordinates $(x, y, z)$ for each pixel using depth image 3. Extract connected regions from segmentation image 4. Filter valid obstacles through distance threshold $dist_{threshold}$ 5. Store obstacle 3D bounding box information #### 3. Collision Risk Assessment Mechanism Calculates collision risk based on safe distance angle $\varphi$ (Equation 11): $$R_{col} = \begin{cases} 1.0 - \frac{\varphi - \theta}{\theta_L - \theta}, & \theta < \varphi < \theta_L \\ 1.0, & \varphi < \theta \end{cases}$$ Where: - $\theta_L, \theta_R$ are two boundary angles - Determines which boundary to use through cross product of $-\overrightarrow{d_o}$ and $\overrightarrow{d_t}$ - $\varphi$ is current flight path angle - $R_{col}$ is collision risk value (between 0-1) #### 4. Auxiliary Sub-target Generation Strategy Dynamically generates auxiliary sub-targets when high collision risk is detected: **Detection Range**: 90° sector region centered at current UAV position (radius $d_{pre}$) **Sub-target Coordinates** (Equations 12-13): $$x_{aux_i} = x_{obs} \pm \sqrt{\frac{d_{safe}^2}{1+ (-\frac{1}{k_1})^2}}, (i = 1,2)$$ $$y_{aux_i} = y_{obs} \pm \sqrt{\frac{d_{safe}^2}{1+ (-\frac{1}{k_1})^2}}, (i = 1,2)$$ Where: - $L_1$ is obstacle connection line (slope $k_1$) - $L_2$ is line perpendicular to $L_1$ - $d_{safe}$ is safe distance - Two auxiliary target points are determined by intersection of $L_2$ and safe distance circle #### 5. Improved Artificial Potential Field Functions **Enhanced Attractive Force** (Equations 14-15): $$F_{en\_att}(target) = k_{att}e^\rho$$ $$\rho = \frac{||P_s - P_t||^\gamma}{||P_s - P_t||^2 + ||P_{UAV\_cur} - P_t||}$$ Where: - Introduces ratio of start-to-target distance and current position-to-target distance - $\gamma$ is adjustment coefficient - Provides stronger pulling force when approaching target, ensuring precise arrival **Auxiliary Sub-target Attractive Force** (Equation 16): $$F_{aux\_att}(target_{aux}) = k_{att}e^{(\frac{\delta}{||P_{UAV\_cur}-P_{aux}||})^2}$$ Where $\delta$ is distance parameter **Total Attractive Force** (Equation 17): $$F_{att} = \begin{cases} F_{en\_att}(target), & \nexists(target_{aux}) \\ F_{aux\_att}(target_{aux}), & \exists(target_{aux}) \end{cases}$$ **Repulsive Force** (Equation 9, traditional APF): $$F_{rep}(p) = \begin{cases} k_{rep} (\frac{1}{d(p, p_o)} - \frac{1}{d_o})\frac{1}{d^2(p, p_o)} \overrightarrow{d_o}, & d(p, p_o) \leq d_o \\ 0, & d(p, p_o) > d_o \end{cases}$$ **Total Force** (Equation 10): $$F(p) = F_{att}(p) + \sum_{i=1}^{n}F_{rep_i}(p) + F_{int}$$ ### Technical Innovation Points 1. **Gradient Continuity Optimization**: Improves repulsive potential field function to ensure first-order gradient continuity, avoiding force field mutations 2. **Multi-level Force Field Integration**: Systematically integrates target attraction, obstacle repulsion, and inter-UAV interaction forces into unified potential field framework 3. **Dynamic Risk Assessment**: Real-time collision risk calculation based on flight trajectory angle, more accurate than traditional distance-only methods 4. **Adaptive Sub-targets**: Dynamically generates auxiliary target points based on obstacle relative positions, more flexible than fixed strategies 5. **Enhanced Attractive Force Model**: Improves attractive force calculation through exponential functions and distance ratios, addressing traditional APF's weak attraction when approaching targets ## Experimental Setup ### Dataset **Simulation Environment**: Static obstacle scenarios constructed in AirSim platform - Scenario Type: Unknown static obstacle environment - UAV Quantity: 1, 3, 5 units (multiple experimental configurations) - Obstacles: Multiple irregularly distributed static obstacles ### Evaluation Metrics 1. **Total Path Length** $P_{length}$: $$P_{length} = \sum_{i=1}^{m-1} \sqrt{(x_{n+1} - x_n)^2 + (y_{n+1} - y_n)^2}$$ Where $m$ is flight steps, $n$ is number of UAVs 2. **Heading Angle Change** $A_{change}$: $$A_{change} = \frac{180° × arctan(\frac{y_{n+1} - y_n}{x_{n+1} - x_n})}{\pi}$$ 3. **Heading Angle Mutation Frequency** $A_{change\_count}$: $$A_{change\_count} = \sum_{i=1}^{n-1} I(A_{change} > 5°)$$ Where $I$ is indicator function, incrementing by 1 when angle change exceeds 5° ### Comparison Methods 1. **T-APF**: Traditional Artificial Potential Field method 2. **I-APF**: Improved method from reference [28] 3. **O-APF**: Hybrid optimization method proposed in this paper ### Implementation Details **Key Parameter Settings** (Table 1): - Communication radius $r = 6$ - Ideal formation distance $d = 4$ - Distance deviation threshold $\varphi = 0.7$ - Interaction force coefficients $\alpha = 0.1, \beta = 10$ - Attractive force coefficient $k_{att} = 1$ - Repulsive force coefficient $k_{rep} = 500$ - Prediction distance $d_{pre} = 6$ - Adjustment coefficients $\gamma = 1.2, \delta = 16$ - Distance threshold $dist_{threshold} = 6$ ## Experimental Results ### Main Results #### 1. Path Smoothness Comparison (Figure 8) Trajectory comparison of three methods in 5-UAV scenario: - **T-APF** (Fig 8.a): Path exhibits obvious irrational detours and sharp turns - **I-APF** (Fig 8.b): Improved path smoothness, but irrational avoidance caused by inter-UAV interaction forces remains - **O-APF** (Fig 8.c): Smoothest path, effectively resolves irrational behavior caused by inter-UAV interaction forces **Key Finding**: Auxiliary sub-target generation mechanism positively mitigates irrational behaviors in multi-UAV formation avoidance; optimizing attractive potential field is more effective than solely improving repulsive potential field for enhancing formation avoidance performance. #### 2. Heading Angle Change Comparison (Figure 9) - **T-APF** (Fig 9.a): Frequent and large heading angle changes - **I-APF** (Fig 9.b): 32% reduction in heading angle change frequency - **O-APF** (Fig 9.c): 84% reduction in heading angle change frequency **Quantitative Result**: O-APF significantly outperforms I-APF in optimizing path smoothness. #### 3. Quantitative Performance Comparison (Table 2) | Algorithm | Path Length [m] (1/3/5 UAVs) | Improvement Rate | Heading Changes [N] | Improvement Rate | |-----------|------------------------------|------------------|-------------------|------------------| | T-APF | 49.26/307.27/805.56 | - | 98/214/290 | - | | I-APF | 47.58/302.61/786.36 | 3.5%/1.5%/2.4% | 60/162/219 | 63%/32%/32% | | O-APF | 46.18/300.31/774.53 | 6.6%/2.3%/4.0% | 18/82/157 | 440%/160%/84% | **Key Observations**: - O-APF outperforms I-APF and T-APF on all metrics - Improvement magnitude decreases with increasing UAV quantity, but maintains performance advantage - Path length improvement exhibits non-linear decay - Heading stability improvement is most significant (84% reduction in 5-UAV scenario) ### Ablation Study While the paper lacks explicit ablation study section, indirect verification through I-APF comparison reveals: - **Auxiliary Sub-target Mechanism**: Both I-APF and O-APF include this mechanism, significantly improving path smoothness compared to T-APF - **Optimized Attractive Potential Field**: Unique to O-APF, further enhancing performance compared to I-APF, particularly in handling inter-UAV interaction force interference - **Collision Risk Assessment**: Synergizes with auxiliary sub-target strategy for dynamic avoidance decision-making ### Experimental Findings 1. **Synergistic Effect**: Combination of auxiliary sub-target generation mechanism and optimized attractive potential field produces significant synergistic effect 2. **Scalability Limitation**: Performance improvement magnitude decreases with increasing UAV quantity, suggesting potential challenges in large-scale swarm scenarios 3. **Heading Stability Advantage**: O-APF's heading stability improvement (84%) far exceeds path length improvement (4%), indicating algorithm's particular effectiveness in enhancing flight smoothness 4. **Practical Validation**: Algorithm effectively operates in unknown obstacle environments, satisfying real-time requirements of practical applications ## Related Work ### 1. Path Planning Method Classification Paper surveys mainstream path planning approaches: - **Search-based Methods**: A* algorithm [5] - **Heuristic Optimization Algorithms**: Genetic algorithms [6], Particle Swarm Optimization [7], Ant Colony Optimization [8] - **Sampling-based Methods**: RRT algorithm [9] - **Velocity Obstacle Method**: VO method [10,11] - **Learning-based Methods**: Reinforcement learning [12] - **Potential Field Methods**: APF methods [13-21] ### 2. APF Method Improvement Research **Obstacle Threat Assessment**: - Feng et al. [14]: Incorporates velocity constraints into repulsive force calculation, quantifying obstacle threat levels **Local Minimum Problem**: - Yang et al. [15]: Optimizes attractive force model, introducing auxiliary target points and assessment mechanisms - **Limitation**: Does not dynamically set auxiliary sub-targets based on relative position information **Path Length Optimization**: - Jiang et al. [20]: Adaptive step-size adjustment strategy - Li et al. [21]: Introduces ratio of total path length to current distance for dynamic step-size adjustment ### 3. Multi-robot Formation Control - Graph theory-based formation modeling [23,24] - Distributed multi-agent coordination [24] ### 4. Advantages of This Work Compared to Related Work - **Systematic Integration**: First to systematically integrate multi-robot formation control and single-machine path optimization mechanisms - **Dynamic Risk Assessment**: Collision risk assessment based on flight trajectory angle more accurate than traditional distance-based methods - **Enhanced Adaptivity**: Auxiliary sub-target strategy dynamically generated based on obstacle relative positions, more flexible than fixed strategies - **Multi-objective Optimization**: Simultaneously optimizes path length and heading stability, better aligning with practical application requirements ## Conclusions and Discussion ### Main Conclusions 1. **Problem Resolution**: Successfully addresses three major problems of traditional APF methods in multi-UAV formation avoidance: path redundancy, heading mutations, and inter-UAV collisions 2. **Performance Improvement**: Compared to traditional T-APF method, O-APF improves path length by 4.0% and reduces heading angle change frequency by 84% (5-UAV scenario) 3. **Mechanism Verification**: - Collision risk assessment mechanism effectively identifies high-risk situations - Auxiliary sub-target strategy significantly improves path smoothness - Optimized attractive potential field ensures precise target arrival 4. **Applicability**: Algorithm demonstrates good applicability and effectiveness in unknown static obstacle environments ### Limitations 1. **Environmental Constraints**: - Validated only in static obstacle environments, not considering dynamic obstacles - Complex 3D environments untested 2. **Scalability**: - Performance improvement magnitude decreases with increasing UAV quantity - Large-scale swarm scenario performance insufficiently verified 3. **Practical Constraints**: - Does not consider wind disturbances, communication delays, and other practical factors - Performance on real hardware platforms unverified 4. **Computational Complexity**: Paper lacks analysis of algorithm computational complexity and real-time performance 5. **Parameter Sensitivity**: Involves multiple parameters (10+), with insufficient guidance for parameter tuning and sensitivity analysis ### Future Directions Research directions explicitly proposed in paper: 1. **Potential Field Optimization**: Further research into potential field theory, optimizing attractive and repulsive potential fields for better suitability to UAV path planning and obstacle avoidance 2. **Strategy Improvement**: Further optimize auxiliary target point avoidance strategy, improving robot avoidance efficiency and shortening avoidance path **Potential Extension Directions**: 3. Dynamic obstacle scenario adaptation 4. Three-dimensional complex environment path planning 5. Large-scale UAV swarm optimization 6. Real hardware platform verification 7. Sensor fusion integration (e.g., LiDAR) ## In-depth Evaluation ### Strengths 1. **Clear Problem Definition**: Accurately identifies three core problems of traditional APF in multi-UAV scenarios, with explicit research motivation 2. **Reasonable Method Design**: - Systematically integrates formation control and single-machine optimization, reflecting holistic thinking - Collision risk assessment based on trajectory angle rather than pure distance, better aligning with actual flight characteristics - Cleverly designed auxiliary sub-target strategy with dynamic generation through geometric relationships 3. **Sufficient Experimental Design**: - Compares traditional methods and latest improvements - Multiple UAV quantity configurations (1/3/5 units) - Comprehensive quantitative metrics (path length, heading change frequency) 4. **Significant Results**: 84% heading stability improvement demonstrates practical application value 5. **Clear Writing**: - Rigorous mathematical formula derivation - Rich figures (9 figures, 2 tables) - Complete logical structure ### Shortcomings 1. **Insufficient Theoretical Analysis**: - Lacks convergence proof - No analysis of algorithm stability conditions - Parameter selection lacks theoretical guidance 2. **Experimental Limitations**: - Validated only in simulation environment, lacking real hardware experiments - Static obstacle scenarios overly simplified - Single environment configuration, lacking diversity 3. **Insufficient Comparison**: - No comparison with other mainstream methods (e.g., RRT, reinforcement learning) - Lacks computational time comparison - No statistical significance testing 4. **Scalability Issues**: - Performance improvement decreases with increasing UAV quantity - Many parameters (10+), complex tuning - Large-scale scenario applicability questionable 5. **Limited Innovation**: - Primarily combination and improvement of existing methods - Core mechanisms (auxiliary sub-targets, risk assessment) have similar ideas in related work - Lacks breakthrough theoretical contributions 6. **Missing Details**: - Obstacle detection method description too brief - Unclear how to select between two auxiliary sub-targets - Formation recovery mechanism not detailed ### Impact 1. **Academic Contribution**: - Provides practical solution for multi-UAV formation obstacle avoidance - Validates effectiveness of integrating multi-level force fields and dynamic sub-target strategies - Can serve as reference baseline for subsequent research 2. **Practical Value**: - Significant heading stability improvement important for actual flight control - Low algorithm computational complexity suitable for real-time applications - Potential applications in static environment monitoring, cooperative search scenarios 3. **Reproducibility**: - Complete parameter settings (Table 1) - Detailed mathematical formulas - Uses open-source AirSim simulation platform - **Shortcoming**: No code provided, some implementation details insufficiently detailed 4. **Limitations**: - Limited application scenarios (static environments) - Requires further verification for practical deployment - Less competitive compared to cutting-edge deep learning methods ### Applicable Scenarios **Suitable Applications**: 1. **Static Environment Inspection**: Building inspection, infrastructure monitoring 2. **Cooperative Search**: Area coverage in rescue operations 3. **Agricultural Applications**: Formation spraying, crop monitoring 4. **Environmental Monitoring**: Atmospheric sampling, pollution source localization 5. **Educational Demonstration**: Multi-robot coordination control teaching **Unsuitable Scenarios**: 1. **Highly Dynamic Environments**: Urban traffic, crowded areas 2. **Complex 3D Spaces**: Indoor multi-story buildings, forest environments 3. **Adversarial Scenarios**: Requiring rapid maneuvering and unpredictable trajectories 4. **Large-scale Swarms**: Tens or more UAVs coordination 5. **Extreme Real-time Requirements**: High-speed flight avoidance **Application Recommendations**: - As basic avoidance module combined with other advanced planning algorithms - Local optimization of global path planning in known map environments - Integration with learning-based methods, using APF to provide safety constraints ## References **Key Cited Works**: 1. **[1] Shao et al. 2024**: Dynamic Obstacle-Avoidance Algorithm for Multi-Robot Flocking Based on Improved APF (main foundational work MRF-IAPF) 2. **[14] Feng et al. 2021**: Incorporates velocity constraints into repulsive force calculation, quantifying obstacle threat levels 3. **[15] Yang et al. 2021**: Optimizes attractive force model, introducing auxiliary target points to reduce local minima 4. **[20] Jiang et al. 2022**: Adaptive step-size adjustment strategy optimizing path length 5. **[21] Li et al. 2022**: Dynamic adaptive step-size adjustment method 6. **[27] Hao et al. 2023**: Single-UAV improved APF method (verification basis for some mechanisms) 7. **[28] Zhuang & Zhu 2024**: APF improvement based on triangular target strategy (comparison method I-APF) --- **Overall Assessment**: This paper addresses practical problems in multi-UAV formation obstacle avoidance, proposing a systematic APF improvement scheme. The method design is reasonable, experimental results are significant, particularly achieving 84% improvement in heading stability with good practical value. However, the paper has room for improvement in theoretical depth, experimental diversity, and innovation, particularly requiring further verification in real hardware platforms and more complex scenarios. Suitable as practical reference for multi-UAV coordination control field, but some distance remains from breakthrough academic contributions.