Pith. sign in

REVIEW 27 references

RF-Source Seeking with Obstacle Avoidance using Real-time Modified Artificial Potential Fields in Unknown Environments

T0 review · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A drone using only a radio bearing reaches targets 46% more often in unknown maps.

desk verdict The APF sampling idea is worth a look, but the paper's 360° AoA claim collapses because the half-plane test in Algorithm 1 cannot decide left from right. read the letter →

arxiv 2506.06811 v1 pith:ANV4OUZV submitted 2025-06-07 cs.RO

Shahid Mohammad Mulla , Aryan Kanakapudi , Lakshmi Narasimhan , Anuj Tiwari This is my paper
classification cs.RO
keywords UAVnavigationRFsourceseekingangleofarrivalartificialpotentialfieldsobstacleavoidanceunknownenvironmentssampling-basedpathplanningGazebosimulation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that a drone can chase a radio transmitter through an unknown, static obstacle field using only the bearing to the transmitter, not its coordinates and not a map that was tuned in advance. The authors combine an angle-of-arrival estimator that recovers 360 degrees of direction from phase differences across four antennas with an artificial potential field navigator whose parameters are re-sampled and cost-weighted in real time after every few waypoints. In Gazebo/ROS2 simulations the bearing estimator has an average angular error of 1.48 degrees, and the adaptive navigator raises the success rate of reaching the target by 46 percent while trimming average trajectory length by 1.2 percent relative to a standard potential field. If these numbers hold, a search-and-rescue drone could home on a cell phone or beacon without GPS, without emitter coordinates, and without re-tuning the planner for each new obstacle layout.

What carries the argument

Two mechanisms carry the argument. The first is the square four-antenna AoA configuration: it turns two ambiguous per-dipole arrival angles $\theta_1,\theta_2$ into one full-360-degree bearing $\theta_0$ through the cotangent identity $\cot\theta_0 = (\cot\theta_1+\cot\theta_2)/2$, with the half-plane fixed by the sign of $\theta_1-\theta_2$; when the two direction rays are parallel the algorithm falls back to either ray until the geometry is non-singular again. The second is an MPPI-inspired (Model Predictive Path Integral) parameter sampler for APFs: at each waypoint it samples new values of $k_{\mathrm{att}}$, $k_{\mathrm{rep}}$, and $d_0$, generates a potential-field trajectory for each sample, weights trajectories by $\exp(-J/\lambda)$ using the cost $J$, and selects the sampled trajectory closest to the weighted average. This machinery replaces per-map manual tuning with a short stochastic search executed on every horizon, which is what lets a single planner cross regions of sharply different obstacle density.

What would settle it

Run the same square-antenna estimator in a scene with the RF source placed behind an obstruction or with a reflective wall nearby, compare the estimated bearing against the true bearing from odometry, and record the mean error; if the estimate flips by roughly 180 degrees or the mean error exceeds 1.48 degrees whenever line of sight is broken, the central sensing claim fails.

Watch

Extended reading notes

Core claim

The paper claims that RF-source seeking and obstacle avoidance can be joined into one bearing-only planner that works across environments it has never seen. For sensing, it derives an angle-of-arrival estimator on a square four-antenna array and resolves the usual 180-degree ambiguity by comparing the two measured angles: the source half-plane is chosen by whether $\theta_1$ or $\theta_2$ is larger, and the final bearing is set by $\cot\theta_0 = (\cot\theta_1 + \cot\theta_2)/2$. For planning, it treats the attractive coefficient, the repulsive coefficient, and the obstacle influence distance of a log-repulsive artificial potential field as Gaussian random variables, samples parameter sets, computes one gradient-descent trajectory for each set, scores the trajectories on length, terminal error, angle changes, and obstacle proximity, and then combines them with cost-weighted averaging instead of trusting any single sample. The reported outcome is that the bearing estimate is accurate to an average of 1.48 degrees and that the adaptive planner reaches the target successfully in 46 percent more runs than a standard potential field in five maps of different obstacle density, with an average trajectory-length reduction of 1.2 percent.

Load-bearing premise

The entire accuracy claim rests on the radio wave reaching the drone by a single clean line-of-sight path, so the measured phase delays point unambiguously at the source; once reflections, obstructions, or noise break that assumption, the half-plane decision can choose the wrong side and the reported error and success-rate gains do not transfer.

Editorial extensions

If this is right

  • A drone carrying this stack can home on an unseen radio emitter using only bearing information, which covers disaster-response scenarios where the target is a phone or beacon and GPS is unavailable.
  • The planner no longer needs a tuned potential-field parameter set for each map; the same initial parameters are re-sampled and reweighted as obstacle density changes, which is what the fifth map is designed to test.
  • Because the target input is only a direction vector, the navigation core composes with any local mapping or SLAM module that can output obstacle circles.
  • At the reported 1.48-degree mean bearing error, a source ten meters away is uncertain by roughly 0.26 meters laterally, so the final landing-pad visual sequence remains necessary to close the last gap.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same cotangent-averaging identity is stated for an arbitrary number of arrays as a least-squares problem, so the sensing core could be extended to more antennas or a three-antenna layout without touching the navigation stack.
  • The half-plane rule $\theta_1 > \theta_2$ is the most fragile link: under multipath the phase difference can flip by half a wavelength, so a natural extension is to fuse this bearing estimate with RSSI or with a second, physically separated array to resolve the side without relying on fine phase comparison.
  • The paper does not adapt the sampling variance or sample count to local obstacle density; an adaptive rule that increases samples where density variance is high is a testable improvement directly suggested by Map 5's higher variance.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the AoA bearing estimate and the MPPI-style APF parameter sampling follow an independent derivation chain from measured phases and cost-based selection; the map-2-tuned starting parameters are a shared initial condition, and the line-of-sight / half-plane issues are validity caveats, not circular reductions.

full rationale

The paper's derivation chain is self-contained at every load-bearing step. RF bearing estimation: θ1 and θ2 are measured from simulated phase differences via Eq. (2); the source coordinates (sx, sy) solve the linear system Ax = b (Eqs. 16, 27) built from antenna geometry and the measured AoAs; the direction-vector formula cot θ0 = (cot θ1 + cot θ2)/2 (Eqs. 29–30) is independently cross-checked by the sine rule (Eqs. 31–32). The true source position is never used to set parameters or disambiguate angles, so the reported 1.48° error is a measured simulation outcome, not a fitted one. Navigation: the modified APF samples potential-field parameters from Gaussians (Eqs. 38–40), scores trajectories with an explicit cost function (Eqs. 41–45), and selects a weighted-optimal trajectory (Eqs. 46–47); success rates over five maps are stochastic measurements, not quantities forced by construction. Caveats that do not rise to circularity: (i) Section V-B states 'The initial parameters for all these runs are set to be the parameters obtained by experimentally tuning the standard APF to work on map 2,' so baseline and proposed method share a hand-fitted starting point; because the baseline is fixed at its best-case map while the modified APF must still search, and still fails 43% of Map-5 runs, the reported gains are not statistically forced by the fit. (ii) The Conclusion concedes the RF algorithm 'only works for the line of sight scenarios; if there is no line of sight, the drone will be confused about which angle reading to consider,' a validity limit rather than a circular reduction. (iii) The Algorithm 1 half-plane rule ('if θ1 > θ2 then Source lies in the half-plane of A1A2') and Appendix B's dismissal of negative k12/k34 values are correctness risks a reviewer should verify, but they are mathematical-validity concerns, not equivalence-by-construction. No load-bearing self-citations exist; all cited methods are external (AoA [9], MUSIC [11], MPPI [23], log-repulsive potential [27]).

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The claimed system behavior rests on hand-tuned navigation hyperparameters and idealized RF and obstacle assumptions. No real RF hardware or non-line-of-sight test was performed.

free parameters (5)
  • Initial APF parameters (k_att, k_rep, d_0) = not reported (tuned experimentally for Map 2)
    Used as sampling means for the modified APF and as fixed values for the standard APF baseline; the comparison and adaptation depend on this hand-tuned starting point.
  • Sampling standard deviations (sigma_att, sigma_rep, sigma_d0) = not reported
    Control the exploration range of MPPI-style parameter sampling; no sensitivity analysis, so the reported success rates may depend on these choices.
  • Weight decay factor lambda = not reported
    Sets how strongly low-cost trajectories dominate the weighted average in Eq. (46); no ablation.
  • Cost function weights (implicit 1 for L, E, A, P) = 1 each (implicit)
    Eq. (41) sums length, goal error, angle deviation, and inverse proximity without normalization; dimensional inconsistency and lack of tuning may affect the selected trajectory.
  • Temporary target distance, waypoint count, sampling count, grid resolution = 2 m, 15 waypoints, update at waypoint 5, N_sample=10, 0.5 m grid
    Hand-set simulation parameters that govern how often the potential field is recomputed and how many samples are used; directly affect success and path length.
assumptions (5)
  • domain assumption The RF source is a point emitter in the same 2D plane as the drone, and the received signal is a clean single-path wave W(x,y,z,t) in Eq. (48) with no multipath.
    The AoA derivation and half-plane disambiguation in Algorithm 1 require an unambiguous phase difference; the authors note in the conclusion that non-line-of-sight breaks the estimate.
  • domain assumption All obstacles can be represented as static circles in the XY plane by fitting a circle to the point-cloud slice at drone altitude.
    Used throughout the simulation and the repulsive potential (Eqs. 34-36); the paper says only circular obstacles were tested.
  • domain assumption Gradient descent on the discrete potential field from the drone position reaches the temporary target and avoids obstacles.
    The algorithm assumes each sampled trajectory is collision-free because it comes from gradient descent; this is not guaranteed with discrete steps and local minima (Section III.B.3 overclaims a guarantee).
  • domain assumption The temporary target at 2 m along the bearing is a valid proxy for the true goal and updating it at waypoint 5 preserves convergence.
    The navigation loop (Algorithm 2) relies on this receding-horizon structure; no proof of convergence is provided.
  • domain assumption Azimuth-only bearing is sufficient; source elevation is ignored.
    The derivation uses only x-y coordinates and Eq. (48) includes z, but the AoA estimation effectively assumes source and antennas at the same height.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RF-Source Seeking with Obstacle Avoidance using Real-time Modified Artificial Potential Fields in Unknown Environments." pith.science (2026). https://pith.science/paper/ANV4OUZV

@misc{pith2026250606811,
  author       = {Pith},
  title        = {Pith review of: RF-Source Seeking with Obstacle Avoidance using Real-time Modified Artificial Potential Fields in Unknown Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ANV4OUZV}},
  note         = {Machine review of arXiv:2506.06811}
}
read the original abstract

Navigation of UAVs in unknown environments with obstacles is essential for applications in disaster response and infrastructure monitoring. However, existing obstacle avoidance algorithms, such as Artificial Potential Field (APF) are unable to generalize across environments with different obstacle configurations. Furthermore, the precise location of the final target may not be available in applications such as search and rescue, in which case approaches such as RF source seeking can be used to align towards the target location. This paper proposes a real-time trajectory planning method, which involves real-time adaptation of APF through a sampling-based approach. The proposed approach utilizes only the bearing angle of the target without its precise location, and adjusts the potential field parameters according to the environment with new obstacle configurations in real time. The main contributions of the article are i) an RF source seeking algorithm to provide a bearing angle estimate using RF signal calculations based on antenna placement, and ii) a modified APF for adaptable collision avoidance in changing environments, which are evaluated separately in the simulation software Gazebo, using ROS2 for communication. Simulation results show that the RF source-seeking algorithm achieves high accuracy, with an average angular error of just 1.48 degrees, and with this estimate, the proposed navigation algorithm improves the success rate of reaching the target by 46% and reduces the trajectory length by 1.2% compared to standard potential fields.

Figures

Figures reproduced from arXiv: 2506.06811 by the authors.

Figure 1
Figure 1. Angle of Arrival in case of 2 antennas. A common implementation uses a Uniform Linear Array (ULA), where the nth antenna receives the incoming signal: sn(t) = A · e j(ωt+nϕ) + w(t) (1) Here, A is amplitude, ω is the carrier frequency, ϕ is the AoA-induced phase shift, and w(t) the noise. To extract the phase, the received signals are correlated with a local carrier frequency of ω, digitized and processed to get ϕ. T… view at source ↗
Figure 3
Figure 3. Arbitrary Antenna Configuration with 4 Antennas. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. If an AoA, θ, is detected, it remains ambiguous whether the source resides on the positive or negative side of the y-axis. In other words, the ambiguity arises between two potential AoA values, θ1 and θ2, which, while mathematically equivalent, correspond to distinct source positions [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Two Angles of Arrival for the dipole A1A2 & A3A4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Square Antenna Configuration. The square configuration is specifically chosen as it repre￾sents the primary proposal for the antenna arrangement due to its symmetry, computational simplicity & simplification of implementation of the algorithm. Referring to [PITH_FULL_…
Figure 6
Figure 6. Figure 6: An example potential function. Till now, it is assumed that the location of all obstacles and the final goal are known from the start. If the environment is unknown and the drone is mapping the obstacles as it goes, the potential needs to be updated as well. When new o…
Figure 7
Figure 7. Figure 7: Drone trajectory during RF source seeking. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 9
Figure 9. Figure 9: Success rates of standard APF and modified APF in [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 11
Figure 11. Figure 11: figure 11. This can be verified by the low variance of obstacle [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 10
Figure 10. Figure 10: Average relative lengths of standard APF and modified [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 12
Figure 12. Figure 12: Map with varying obstacle density. line of sight scenarios; if there is no line of sight, the drone will be confused about which angle reading to consider due to real-world factors like noise and multipath effects. Enhancing robustness using more antennas and estimati…
Figure 13
Figure 13. Figure 13: No Solution Case when det(A) = 0. In the case where no solution exists, the direction vectors corresponding to the AoA at A1A2 and A3A4 are parallel in the drone body coordinate system, as depicted in Figure [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]
Figure 15
Figure 15. Figure 15: Diverging Direction vectors at A12 & A34. amount of noise is present, in which case techniques like MUSIC algorithm should be implemented before this to filter out the noise. Thus, the presence of negative scaling factors k12 or k34 can be disregarded as a practical c…
Figure 14
Figure 14. Figure 14: Infinite Solution Case when det(A) = 0. In the second case, where infinitely many solutions exist, the direction vectors remain parallel; however, the source, the midpoint of A1 and A2, and the midpoint of A3 and A4 are collinear. This configuration is illustrated in …
Figure 16
Figure 16. Figure 16: An Arbitrary Antenna Configuration with 3 Antennas. [PITH_FULL_IMAGE:figures/full_fig_p013_16.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages

  1. [1]

    Wo ´zniak

    Sven Mayer, Lars Lischke, and Pawel W. Wo ´zniak. Drones for Search and Rescue. In 1st International Workshop on Human-Drone Interaction, Glasgow, United Kingdom, May 2019. Ecole Nationale de l’Aviation Civile [ENAC]

    work page 2019

  2. [2]

    Drone-based bridge inspections: Current practices and future directions

    Tommaso Panigati, Mattia Zini, Domenico Striccoli, Pier Francesco Giordano, Daniel Tonelli, Maria Pina Limongelli, and Daniele Zonta. Drone-based bridge inspections: Current practices and future directions. Automation in Construction , 173:106101, 2025

    work page 2025

  3. [3]

    Using smartphones to locate trapped victims in disasters

    Yenpo Tai and Teng-To Yu. Using smartphones to locate trapped victims in disasters. Sensors (Basel), 22(19):7502, October 2022

    work page 2022

  4. [4]

    Au- tonomous multicopter landing on a moving vehicle based on rssi

    Jongwoo An, Hosun Kang, Jiwook Choi, and Jangmyung Lee. Au- tonomous multicopter landing on a moving vehicle based on rssi. Journal of Navigation , 76(4–5):545–561, 2023

    work page 2023

  5. [5]

    A survey on radio frequency based precise localisation technology for uav in gps-denied environment

    Beiya Yang and Erfu Yang. A survey on radio frequency based precise localisation technology for uav in gps-denied environment. Journal of Intelligent & Robotic Systems , 103(3), October 2021

    work page 2021

  6. [6]

    Faheem Zafari, Athanasios Gkelias, and Kin K. Leung. A survey of indoor localization systems and technologies. IEEE Communications Surveys & Tutorials, 21(3):2568–2599, 2019

    work page 2019

  7. [7]

    Rssi- based localisation of the radio waves source by mobile agents

    Dariusz Czerwinski, Slawomir Przylucki, and Dmitry Mukharsky. Rssi- based localisation of the radio waves source by mobile agents. In Piotr Gaj, Andrzej Kwiecie ´n, and Piotr Stera, editors, Computer Networks , pages 370–383, Cham, 2016. Springer International Publishing

    work page 2016

  8. [8]

    A generalized tdoa/toa model for tof positioning

    Maximilian von Tschirschnitz, Marcel Wagner, Marc-Oliver Pahl, and Georg Carle. A generalized tdoa/toa model for tof positioning. In 2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), page 1–8. IEEE, September 2019

    work page 2019

Show all 27 references
  1. [9]

    Rf angle of arrival-based node localisation

    Isaac Amundson, J ´anos Sallai, Xenofon Koutsoukos, Akos Ledeczi, and Miklos Maroti. Rf angle of arrival-based node localisation. IJSNet, 9:209–224, 05 2011

    2011

  2. [10]

    Wi-fi based indoor positioning system for mobile robots by using particle filter

    Hikmet Yucel, Gulin Elibol, and Ugur Yayan. Wi-fi based indoor positioning system for mobile robots by using particle filter. CoRR, abs/2012.05286, 2020

    2012 arXiv

  3. [11]

    R. Schmidt. Multiple emitter location and signal parameter estimation. IEEE Transactions on Antennas and Propagation, 34(3):276–280, 1986

    1986

  4. [12]

    Tarig Ballal and C. J. Bleakley. 3d location and orientation estimation using angle of arrival. In 2009 IEEE International Symposium on Intelligent Signal Processing , pages 21–26, 2009

    2009

  5. [13]

    C.W. Warren. Global path planning using artificial potential fields. In Proceedings, 1989 International Conference on Robotics and Automa- tion, pages 316–321 vol.1, 1989

    1989

  6. [14]

    An advanced potential field method proposed for mobile robot path planning

    Ameni Azzabi and Khaled Nouri. An advanced potential field method proposed for mobile robot path planning. Transactions of the Institute of Measurement and Control , 41(11):3132–3144, 2019

    2019

  7. [15]

    Koren and J

    Y . Koren and J. Borenstein. Potential field methods and their inherent limitations for mobile robot navigation. In Proceedings. 1991 IEEE International Conference on Robotics and Automation, pages 1398–1404 vol.2, 1991

    1991

  8. [16]

    A new technique to escape local minimum in artificial potential field based path planning

    Min Gyu Park and Min Cheol Lee. A new technique to escape local minimum in artificial potential field based path planning. KSME International Journal, 17(12):1876–1885, December 2003

    2003

  9. [17]

    Artificial potential field based path planning for mobile robots using a virtual obstacle concept

    Min Cheol Lee and Min Gyu Park. Artificial potential field based path planning for mobile robots using a virtual obstacle concept. In Proceedings 2003 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM 2003) , volume 2, pages 735–740 vol.2, 2003

    2003

  10. [18]

    Mora and Josep Tornero

    Marta C. Mora and Josep Tornero. Path planning and trajectory generation using multi-rate predictive artificial potential fields. In 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems , pages 2990–2995, 2008

    2008

  11. [19]

    Improved potential field method path planning based on genetic algo- rithm

    Feng Liu, Hualing He, Zhihua Li, Zhi-Hong Guan, and Hua O Wang. Improved potential field method path planning based on genetic algo- rithm. In 2020 39th Chinese Control Conference (CCC) , pages 3725– 3729, 2020

    2020

  12. [20]

    Vadakkepat, Kay Chen Tan, and Wang Ming-Liang

    P. Vadakkepat, Kay Chen Tan, and Wang Ming-Liang. Evolutionary artificial potential fields and their application in real time robot path planning. In Proceedings of the 2000 Congress on Evolutionary Computation. CEC00 (Cat. No.00TH8512) , volume 1, pages 256–263 vol.1, 2000

    2000

  13. [21]

    Elkilany, A

    Basma Gh. Elkilany, A. A. Abouelsoud, Ahmed M. R. Fathelbab, and Hiroyuki Ishii. Potential field method parameters tuning using fuzzy inference system for adaptive formation control of multi-mobile robots. Robotics, 9(1):10, February 2020

    2020

  14. [22]

    Optimization of potential field method parameters through networks for swarm cooperative manipulation tasks

    Rocco Furferi, Roberto Conti, Enrico Meli, and Alessandro Ridolfi. Optimization of potential field method parameters through networks for swarm cooperative manipulation tasks. International Journal of Advanced Robotic Systems , 13(6), November 2016

    2016

  15. [23]

    Rehg, and Evangelos A

    Grady Williams, Paul Drews, Brian Goldfain, James M. Rehg, and Evangelos A. Theodorou. Aggressive driving with model predictive path integral control. In 2016 IEEE International Conference on Robotics and Automation (ICRA), pages 1433–1440, 2016

    2016

  16. [24]

    A model predictive path integral method for fast, proactive, and uncertainty-aware uav planning in cluttered environments, 2023

    Jacob Higgins, Nicholas Mohammad, and Nicola Bezzo. A model predictive path integral method for fast, proactive, and uncertainty-aware uav planning in cluttered environments, 2023

    2023

  17. [25]

    Phase-difference measurement-based angle of arrival estimation using long-baseline interferometer

    Van-Sang Doan, Thien Huynh-The, Van-Phuc Hoang, and Jiri Vesely. Phase-difference measurement-based angle of arrival estimation using long-baseline interferometer. IET Radar Sonar Navig. , 17(3):449–465, March 2023

    2023

  18. [26]

    Correlation-based radio localization in an indoor environment

    Thomas Callaghan, Nicolai Czink, Francesco Mani, Arogyaswami Paulraj, and George Papanicolaou. Correlation-based radio localization in an indoor environment. EURASIP J. Wirel. Commun. Netw., 2011(1), December 2011

    2011

  19. [27]

    Enhanced artificial potential field for mass’s path planning navigation in restricted waterways

    Zongkai Wang and Namkyun Im. Enhanced artificial potential field for mass’s path planning navigation in restricted waterways. Applied Ocean Research, 149:104052, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.