{"id":"857012fb-2975-4758-b9ff-46bc85d2c92b","arxiv_id":"2501.09338","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"A modified artificial potential field algorithm with a simulated annealing escape loop is proposed for 3D UAV obstacle avoidance, supported only by qualitative MATLAB simulations.","lead":"This paper combines a force-field navigation method with a random-search refinement loop so a rescue drone can steer around trees, no-fly zones, and moving obstacles in 3D. The proposed planner aims to produce shorter, collision-free routes that escape the dead ends where basic force-field methods stall.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (18) has the wrong sign for the target-directed repulsion term, so the paper's own formula for solving the unreachable-target problem pushes the UAV away from the goal.","rationale":"The abstract's central promise is that the modified field functions overcome APF's unreachable-target defect. The modified repulsion is a standard fix, but the force component implementing the fix must point toward the target. The equations as written point the opposite way. This is not merely a missing convergence proof or an unsupported 'global optimality' phrase; it is a concrete mathematical inconsistency in the formula whose effect the simulations are meant to demonstrate. Since no code, parameter table, or numeric comparison is included, the safest reading is that the manuscript needs a correction and a re-run of the experiments. I therefore keep the reader's CONDITIONAL verdict. The dynamic-obstacle concern raised by the reader is real but secondary: Eq. (4) has no time index and the algorithm does not update obstacle positions, but even static experiments would be affected by the sign issue, and the sign issue attacks the headline improvement directly.","tokens_in":7932,"tokens_out":7993,"duration_ms":76828,"concrete_test":"Symbolically differentiate Eq. (15) to obtain F = −∇Urep with respect to Xu, and check the coefficient of the ∂ρ(Xu,Xg)/∂x term. If the coefficient is negative, implement Eqs. (14)–(18) exactly in the 200m × 200m × 20m scenario with a spherical obstacle placed near the target point; a UAV initialized between the obstacle and the target should be attracted to the target if the sign is correct, and repelled from it if Eq. (18) is used as printed. Report whether the resulting path reaches Xg.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing weakness is an internal inconsistency in the force law that is supposed to solve the unreachable-target problem. In Eq. (15), the repulsive potential Urep includes the factor ρ(Xu,Xg)^μ. Taking the negative gradient of this factor gives a force term −(μ/2)β(1/ρ(Xu,Xoi) − 1/ρ0)^2 ρ(Xu,Xg)^(μ−1) ∇ρ(Xu,Xg), where ∇ρ(Xu,Xg) = (Xu − Xg)/ρ(Xu,Xg) is the unit vector from the target point to the UAV. Eq. (18) instead writes Frepg2 with a plus sign, so as printed Frepg2 points away from the target point, not toward it. The text explicitly states that Frepg2 points toward the target and uses this term to cancel excessive repulsion near the goal. If the displayed equations are implemented literally, the mechanism for target reachability in Figs. 6 and 7 is reversed, and the claimed improvement over classical APF is not supported by the derivation. Because no parameter values or code are supplied, the reader cannot determine whether the simulations used Eq. (18) as printed or a corrected sign.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript addresses 3D UAV path planning in a forest-firefighting scenario with static cylindrical obstacles, static spherical no-fly zones, and moving spherical obstacles. The authors propose APF-SA, which modifies the attractive potential by adding an exponential term, modifies the repulsive potential by multiplying with a target-distance factor, and invokes simulated annealing to escape local minima. The paper reports simulation comparisons against APF, RRT, and RRT* and claims shorter paths, solution of the unreachable-target problem, and robustness in dynamic environments.","tokens_in":8215,"tokens_out":7061,"duration_ms":68062,"significance":"If the central claims held, the paper would offer a useful incremental improvement to potential-field path planning for UAVs: the problem formulation is clear, the algorithm is described in pseudocode, and comparisons against standard baselines are attempted. The main value is conditional, however, because several displayed equations contain sign, unit, and vector-form errors, no numerical results or parameter values are reported, and the claims of global optimality and dynamic-obstacle handling go beyond the presented evidence. With corrected equations, a quantitative evaluation, and a reproducible parameter set, the method could be a reasonable engineering contribution, but the current manuscript does not substantiate its central claims.","major_comments":[{"comment":"The sign of Frepg2 is inconsistent with the force law derived from Eq. (15). Because grad rho(Xu,Xg) = (Xu - Xg)/rho(Xu,Xg) points from the target point to the UAV, the negative gradient of the target-distance factor gives Frepg2 = -(mu/2) beta (1/rho(Xu,Xoi) - 1/rho0)^2 rho(Xu,Xg)^(mu-1) grad rho(Xu,Xg), which points from the UAV toward the target point. Eq. (18) prints a plus sign, so as written Frepg2 points away from the target. This contradicts the sentence in Section III-B stating that Frepg2 'points toward the target point' and reverses the mechanism claimed to solve the unreachable-target problem in Figs. 6 and 7. The authors must correct the sign and state explicitly which force law was implemented in the simulations.","section":"III-B, Eq. (18)"},{"comment":"Eq. (14) is not a valid force expression. It adds a distance rho(Xu,Xg), which has units of meters, to the dimensionless quantity exp(-(rho(Xu,Xg) - epsilon Delta s)), and the right-hand side is a scalar, not a vector, so it cannot be combined with the vector Frep in Eq. (10). Because Fatt drives the UAV toward the target in every simulation, this dimensional and vector inconsistency affects the validity of all reported trajectories. Please rewrite Eq. (14) as a vector equation with consistent units, for example Fatt(Xu) = -eta [rho(Xu,Xg) + exp(-(rho(Xu,Xg) - epsilon Delta s))] (Xu - Xg)/rho(Xu,Xg), and define the units of epsilon and Delta s.","section":"III-B, Eq. (14)"},{"comment":"The central performance claims are not backed by numerical evidence. The text states that each algorithm was run 100 times and that APF-SA has the shortest optimal and average path, with its worst path slightly longer than that of RRT*, but no table of path lengths, runtimes, variances, or success rates is provided, and no parameter values (eta, beta, rho0, epsilon, mu, alpha, T0, Delta s, Delta l) are reported. Without these data, the comparisons in Fig. 8 are not reproducible and the claimed advantage over APF, RRT, and RRT* cannot be assessed. Please add a quantitative results table and the full parameter set used in the simulations.","section":"Section IV"},{"comment":"The dynamic-obstacle claim is not established by the model. Eq. (1) updates the UAV position with a fixed step under the current resultant force, and Eq. (4) checks collision only at the instantaneous position against static obstacle radii; the moving spheres described in Section II-A are given no velocity or acceleration model, and no prediction or replanning over time is specified. Moreover, because collision is checked only at sampled points, the path segment between consecutive waypoints could intersect an obstacle without being detected. Please provide an explicit dynamic obstacle model, a collision test for path segments, or revise the claims about dynamic environments and Fig. 7.","section":"Section II-A and Section IV"},{"comment":"The claim that APF-SA 'converge[s] to globally optimal solutions' is unsupported. Algorithm 1 uses a random nearby point and the Metropolis acceptance rule in Eq. (19), but no convergence proof, cooling-schedule analysis, or argument that the random perturbations can reach all relevant configurations is given; simulated annealing is a stochastic heuristic and does not guarantee global optimality in finite time. Please replace or qualify this claim, and define E(t_n) and E(t_0) in Eq. (19) in terms of the potential field.","section":"Abstract and Section III-C"}],"minor_comments":[{"comment":"The variable gamma_x is described as an angle, but it is used as a displacement component in the update x' = x + Delta s * gamma_x; it should be defined as a component of the unit direction vector.","section":"Section II-A, Eq. (1)"},{"comment":"The notation (Xu - Xoi)^2 is ambiguous for vectors; the squared Euclidean norm ||Xu - Xoi||^2 should be used.","section":"Section II-A, Eq. (3)"},{"comment":"The symbol S in the constraint S is an element of P is not defined; it should be specified as the sequence of waypoints or the continuous path.","section":"Section II-C, Eq. (7)"},{"comment":"The performance comparison figure needs readable axis labels, units, and error bars or standard deviations so that the claim of comparing 100 runs can be assessed.","section":"Section IV and Fig. 8"},{"comment":"The text states that APF-SA shows advantages in time complexity, but no runtime measurements or complexity analysis are provided; please either add this evidence or remove the claim.","section":"Section IV"}],"recommendation":"major_revision","confidential_remarks":"This manuscript reads like an early-stage draft: the absence of any numerical table, parameter values, or code is unusual for a journal submission, and the sign and unit errors in the core equations raise doubts about whether the displayed equations match the implemented algorithm. The main claims are not yet verifiable, so I would not accept the paper in its current form. If the authors can correct the equations, provide a reproducible quantitative evaluation, and temper the global-optimality and dynamic-obstacle claims, the paper could become a satisfactory engineering contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me cut to the chase. This is a standard APF-SA hybrid for 3D UAV path planning. The authors modify the attractive and repulsive potentials following [15,16] and bolt on a simulated annealing escape loop. That exact combination is new, if minor. The qualitative plots show APF-SA reaching targets that classical APF misses, so the approach is plausible in simple simulated scenes.\n\nThe main problem is a sign error in the force law that is supposed to solve the unreachable-target problem. Eq. (18) writes Frepg2 with a plus sign, but the correct negative gradient of the repulsive potential in Eq. (15) gives a minus sign. As printed, Frepg2 points away from the target, not toward it, contradicting the text. That term is the entire mechanism for canceling repulsion near the goal. Without a corrected sign, the claimed fix for unreachable targets is reversed. Since no code or parameter values are given, we can't tell if the simulations used the printed equation or a corrected one. This is a load-bearing flaw, not a typo you can gloss over.\n\nBeyond that, the evaluation is thin. There are no numeric tables, no error bars, no convergence analysis, and no parameter settings. The global optimality claim from SA is not justified—SA escapes local minima but doesn't guarantee global optimality in this setup. The dynamic obstacles are modeled as moving spheres, but collision checks only use the current position via Eq. (4); obstacle velocity is ignored, so the dynamic-avoidance claim is not established. Eq. (14) also mixes units by adding a distance to a dimensionless exponential. The 'robust' in the title is never defined or measured.\n\nWhat the paper does well: it's clearly written, the algorithm is easy to implement, and the comparison against APF, RRT, and RRT* (even though only qualitative) is a reasonable sanity check. The engineering combination is new in the cited literature, so it may be useful to readers looking for a simple APF variant to adapt.\n\nThe paper is fixable: correct the sign, provide actual numbers, release code, and soften the optimality claims. As it stands, it should not be accepted without that. It deserves a serious referee because the core idea is sound and the flaw is identifiable—but the referee will need to demand the missing evidence. I wouldn't cite it yet.","headline":"Plausible APF-SA hybrid, but the unreachable-target fix has a sign error and the claims outrun the evidence.","tokens_in":8762,"tokens_out":5425,"would_cite":false,"duration_ms":47508,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A hybrid potential-field and simulated-annealing planner claims shorter collision-free 3D paths than APF, RRT, and RRT*.","keywords":["UAV path planning","artificial potential field","simulated annealing","obstacle avoidance","3D environment","forest firefighting","no-fly zones","local minimum escape"],"falsifier":"Run APF-SA in the paper's 200 m × 200 m × 20 m scenario with a single sphere obstacle moving perpendicular to the start–goal line at speeds from 1 m/s to 10 m/s, with step sizes $\\Delta s$ in the paper's range; record whether a collision occurs when the obstacle crosses the UAV's path between two position updates. If the collision rate rises with obstacle speed, the instantaneous-distance check in Eq. (4) plus the SA random point is insufficient; if it stays at zero, the dynamic-obstacle claim holds for those speeds.","tokens_in":7708,"feed_emoji":"🚁","tokens_out":12022,"duration_ms":109539,"temperature":0.7,"pith_summary":"APF-SA (artificial potential field with simulated annealing) is a UAV path planner for 3D emergency-rescue flights. The paper tries to show that the two classic failure modes of the artificial potential field—a navigation method in which the target attracts the UAV and obstacles repel it—namely getting trapped in a local minimum and failing to reach a target near an obstacle, can be cured by reshaping the attractive and repulsive forces and by adding a simulated-annealing escape step, a probabilistic search that accepts occasionally worse moves while temperature cools. The paper claims the planner finds shorter collision-free paths than classical APF, RRT, and RRT* in a forest-fire scenario built from cylindrical, spherical, and moving obstacles, with the same algorithm serving for static and dynamic obstacles. The practical payoff is a rescue UAV that can autonomously reach a fire destination under an energy budget; the paper's stronger claim is that the annealing step lets the path converge toward a globally optimal solution.","feed_headline":"Annealing boost beats APF, RRT, RRT* on path length","feed_subtitle":"Modified potential fields plus simulated annealing keep rescue UAVs out of traps and on course in 3D.","key_machinery":"The central machinery is a modified artificial potential field. An artificial potential field treats the UAV as a particle attracted by the target and repelled by obstacles; APF-SA changes the attractive force $F_{\\mathrm{att}}(X_u) = -\\eta[\\rho(X_u,X_g)+e^{-(\\rho(X_u,X_g)-\\varepsilon\\Delta s)}]$ and the repulsive field $U_{\\mathrm{rep}}(X_u) = \\frac{1}{2}\\beta(\\frac{1}{\\rho(X_u,X_{oi})}-\\frac{1}{\\rho_0})^2 \\rho(X_u,X_g)^\\mu$. The first change makes the falloff near the goal reverse, so the target keeps pulling the UAV; the second scales repulsion down to zero at the goal, making the goal the global minimum of the potential landscape. When the resultant force $F_s = F_{\\mathrm{att}} + F_{\\mathrm{rep}}$ is zero, a simulated-annealing loop with temperature update $T(t)=\\alpha T(t-1)$ and Metropolis acceptance inserts a random point to break the stalemate. These pieces carry the argument that the UAV can escape local minima and reach the target on a shorter path.","core_discovery":"The paper's central claim is that APF-SA removes the two standard failures of classical APF in 3D obstacle fields. The attractive force is modified to grow near the goal (Eq. 14), and the repulsive field is multiplied by a positive power of the distance from the UAV to the target (Eq. 15), so the target becomes the lowest-potential point even when an obstacle sits next to it. When the attractive and repulsive forces cancel exactly, the algorithm generates a random reachable point and accepts or rejects it with the Metropolis criterion, breaking the zero-force equilibrium. In the paper's simulations in a 200 m × 200 m × 20 m space, traditional APF oscillates or stalls while APF-SA reaches the target, and over 100 runs APF-SA's optimal and average path lengths are shorter than those of APF, RRT, and RRT*, with only the worst-case path slightly longer than RRT*. The paper concludes that the method is suitable for static and dynamic obstacle environments in forest firefighting.","pith_inferences":["Editorial inference: the global-optimality claim is not proven analytically; a fair test would compare APF-SA's path-length distribution against RRT* over many random obstacle layouts and annealing schedules.","Editorial inference: because Eq. (4) checks only instantaneous distance with no obstacle velocity, a stronger dynamic-obstacle version would add time-to-collision or velocity obstacles to bound the needed look-ahead.","Editorial inference: the energy constraint in Eq. (5) is not part of the minimization objective, so a direct extension is to minimize a weighted sum of path length and energy use rather than length alone.","Editorial inference: the cylinder/sphere geometry suits forest no-fly zones; urban rescue would need signed-distance fields or convex-shape distance functions, but the APF-SA structure would carry over."],"forward_implications":["APF-SA's optimal and average path lengths are shorter than those of APF, RRT, and RRT* over 100 runs in the paper's 3D forest scenario.","The modified potential functions make the target the lowest-potential point even when an obstacle is near it, eliminating the unreachable-target failure.","The simulated-annealing insertion breaks zero-force local-minimum states that stop classical APF.","The same APF-SA planner is reported to handle static and dynamic obstacles without switching modes.","Shorter reported paths translate to lower energy use under the paper's linear energy-per-unit-length model."],"supporting_citations":[{"why":"Supplies the RRT* algorithm that serves as the main non-APF baseline in the 100-run path-length comparison.","marker":"[4]"},{"why":"Defines the classical artificial potential field that APF-SA starts from and is compared against.","marker":"[8]"},{"why":"Documents the local-minimum and oscillation failures that the simulated-annealing escape is designed to fix.","marker":"[9]"},{"why":"Motivates the hybrid APF-plus-stochastic-search design by combining goal-biased APF with RRT*.","marker":"[10]"},{"why":"Cited for the original APF principle and potential-field force model that the modified functions extend.","marker":"[14]"},{"why":"Supplies the improved attractive-potential technique used to make the target reachable despite nearby obstacles.","marker":"[15]"},{"why":"Supplies improved potential-field force shaping used in the repulsion correction of APF-SA.","marker":"[16]"},{"why":"Introduces the simulated annealing metaheuristic that APF-SA invokes at zero-resultant-force states.","marker":"[17]"},{"why":"Provides the Metropolis acceptance criterion and temperature update used in the SA inner loop.","marker":"[18]"}],"fun_headline_variants":["APF-SA escapes local minima to cut UAV path length in 3D","Rescue UAVs dodge obstacles with annealing-boosted potential fields","Simulated annealing fixes APF traps for shorter rescue drone paths","Hybrid APF-SA outperforms RRT and RRT* for rescue UAVs","Annealing-infused APF clears UAV paths in 3D rescue scenes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The dynamic-obstacle claim rests on the assumption that checking only the UAV's current position against fixed-radius cylinder/sphere obstacles at each step is enough to avoid moving obstacles, since the paper models no obstacle velocity or acceleration.","fun_headline_variants_meta":{"raw":{"variants":["APF-SA escapes local minima to cut UAV path length in 3D","Rescue UAVs dodge obstacles with annealing-boosted potential fields","Simulated annealing fixes APF traps for shorter rescue drone paths","Hybrid APF-SA outperforms RRT and RRT* for rescue UAVs","Annealing-infused APF clears UAV paths in 3D rescue scenes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000584,"raw_usage":{"total_tokens":2717,"prompt_tokens":884,"completion_tokens":1833,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":500,"completion_tokens_details":{"reasoning_tokens":1734}},"tokens_in":500,"tokens_out":1833,"duration_ms":11686,"temperature":1.0,"reasoning_tokens":1734,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:06:03.263643+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run APF-SA in the paper's 200 m × 200 m × 20 m scenario with a single sphere obstacle moving perpendicular to the start–goal line at speeds from 1 m/s to 10 m/s, with step sizes $\\Delta s$ in the paper's range; record whether a collision occurs when the obstacle crosses the UAV's path between two position updates. If the collision rate rises with obstacle speed, the instantaneous-distance check in Eq. (4) plus the SA random point is insufficient; if it stays at zero, the dynamic-obstacle claim holds for those speeds.","supporting_citations":[{"cited_title":"TargetTree-RRT*: Continuous-Curvature Path Planning Algorithm for Autonomous Parking in Complex Environ- ments,","cited_arxiv_id":null,"evidence_quote":"Supplies the RRT* algorithm that serves as the main non-APF baseline in the 100-run path-length comparison."},{"cited_title":"Distributed Motion Planning for Safe Autonomous Vehicle Overtaking via Artificial Potential Field,","cited_arxiv_id":null,"evidence_quote":"Defines the classical artificial potential field that APF-SA starts from and is compared against."},{"cited_title":"Overtaking Path Planning for CA V Based on Improved Artificial Potential Field,","cited_arxiv_id":null,"evidence_quote":"Documents the local-minimum and oscillation failures that the simulated-annealing escape is designed to fix."},{"cited_title":"UA V trajectory planning based on APF-RRT* algorithm with goal-biased strategy,","cited_arxiv_id":null,"evidence_quote":"Motivates the hybrid APF-plus-stochastic-search design by combining goal-biased APF with RRT*."},{"cited_title":"Path Optimization and Obstacle Avoidance using Gradient Method with Potential Fields for Mobile Robot,","cited_arxiv_id":null,"evidence_quote":"Cited for the original APF principle and potential-field force model that the modified functions extend."},{"cited_title":"Obstacle Avoidance for Mi- crorobots in Simulated Vascular Environment Based on Combined Path Planning,","cited_arxiv_id":null,"evidence_quote":"Supplies the improved attractive-potential technique used to make the target reachable despite nearby obstacles."},{"cited_title":"Shared Control in Robot Teleoperation With Improved Potential Fields,","cited_arxiv_id":null,"evidence_quote":"Supplies improved potential-field force shaping used in the repulsion correction of APF-SA."},{"cited_title":"Simulated Annealing Particle Swarm Optimization for High-Efficiency Power Am- plifier Design,","cited_arxiv_id":null,"evidence_quote":"Introduces the simulated annealing metaheuristic that APF-SA invokes at zero-resultant-force states."},{"cited_title":"Performance Optimization of Variable Cycle Engine Based on Improved Simulated Annealing Algorithm,","cited_arxiv_id":null,"evidence_quote":"Provides the Metropolis acceptance criterion and temperature update used in the SA inner loop."}],"review_version":1}