{"id":"1d707b29-08b6-4e5d-976c-e83ae2462e13","arxiv_id":"2506.13106","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A two-drone range-only estimator and anti-synchronization controller lets autonomous guardians encircle and intercept a moving hostile UAV in 3D.","lead":"Two guardian drones use only noisy distance measurements to estimate a hostile drone's position, encircle it, and close in to intercept it. The system is designed for GPS-denied, radar-jammed conditions and is demonstrated with Tello drones in a live overwatch scenario.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The real-world demo computes pseudo ranges from AirSLAM global point clouds, so the claimed range-only interception is not actually tested; the same SLAM output already supplies target positions.","rationale":"The reader's weakest-assumption analysis already identified the pseudo-range validation as the key gap, and I agree that it is the load-bearing weakness of the central claim. My read does not change the verdict: the theory is plausible and the simulation is supportive, but the real-world experiment substitutes camera-derived AirSLAM positions for actual range measurements, so the claimed 'first demonstration' of range-only drone-to-drone interception is not established. The appropriate disposition remains CONDITIONAL: the condition is to validate with a genuine onboard range sensor or to narrow the claimed contribution. I do not see a reason to reject outright, since the estimator/controller may still be sound; the deficiency is in the evidence for the empirical novelty claim, not an internal contradiction in the derivations.","tokens_in":12137,"tokens_out":9572,"duration_ms":118672,"concrete_test":"Re-run the TPE and encirclement controller exactly as in Section IV, but replace the AirSLAM-inferred pseudo ranges with measurements generated from a realistic onboard range sensor model (e.g., UWB or acoustic range = true Euclidean distance plus zero-mean noise with standard deviation 0.3–1 m, 10–20 Hz update, occasional dropouts), while keeping all control parameters fixed. If the 0.05 m estimation error and successful takedown no longer reproduce, the experiment's pseudo-range surrogate is the reason the range-only claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline claim is 'first to demonstrate autonomous drone-to-drone interception using minimal onboard sensors with range-only measurements.' Section IV-B, however, reports no true range sensor. The distances fed to the estimator are 'pseudo range distances ... inferred indirectly' from AirSLAM global point clouds generated by multiple camera-based SLAM instances. This makes the experiment unable to separate the proposed range-only estimation/control loop from a pipeline that already has the hostile drone's 3D position from visual SLAM: the pseudo ranges are deterministic functions of the very SLAM estimates the range-only method is meant to replace. The assertion that camera-derived ranges 'closely resemble real-world measurement uncertainty' is not demonstrated; acoustic or UWB ranging has different noise statistics, multipath, update rates, and outage behavior, and in this setup the errors are correlated with the control loop rather than being independent sensor noise. Consequently, neither the 0.05 m simulation error nor the real-world encirclement trajectories support the range-only claim as stated. A secondary issue is that the velocity compensator in Eq. (11c) is a single impulse at k=km, leaving zero target-velocity compensation for all later steps; this is concerning for continuous target maneuvers, but the pseudo-range substitution alone already makes the central empirical claim unverified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an autonomous 3D target encirclement and interception strategy for two guardian UAVs protecting a surface target from a hostile UAV. The guardians use noisy distance measurements to a non-cooperative target, an RLS-based position estimator with a velocity compensator, and an anti-synchronization (AS) encirclement controller with three concentric decision zones. The paper claims the first demonstration of drone-to-drone interception with range-only measurements, supported by a MATLAB simulation and a real-world experiment using Tello drones. The theoretical contribution is a PE-based estimator and a controller with convergence proofs in Appendices A and B.","tokens_in":12342,"tokens_out":7363,"duration_ms":74263,"significance":"If the convergence proofs were complete and the experiment actually used range-only sensors, the work would be a meaningful step toward lightweight anti-drone systems that operate in GPS-denied and NLOS conditions. The control architecture, combining AS encirclement with vertical jitter and adaptive radius reduction, is interesting, and the simulation shows small estimation and tracking errors. However, the velocity compensator in Eq. (11c) acts only at a single time step, which is not reflected in the proof's assumption of a uniformly bounded velocity error, and the real-world experiment infers pseudo ranges from AirSLAM visual SLAM positions rather than from range sensors. These issues currently prevent the paper from substantiating its central claim.","major_comments":[{"comment":"The velocity compensator in Eq. (11c) is defined with a Kronecker delta δ{k=km}, so v̂2(k)=0 for all k≠km. The proof of Theorem 1 in Appendix A assumes ∥ev,2∥ ≤ ϱ2 for all time, but since the compensation vanishes except at k=km, the error ev,2 = v2 − v̂2 equals the full target velocity v2 for all other steps, yielding only the bound ϱ2 = ˇv2. The resulting estimation error bound ϱ1 = 2t²γ1β̌1ϱ2²/((1−2γ1)β̂1) therefore does not show any benefit of the compensation and does not establish the claimed velocity-compensated convergence for a continuously maneuvering target. The proof must be revised to handle a persistent velocity error or the compensator must be redesigned as a persistent estimator.","section":"Section III-B, Eq. (11c) and Appendix A"},{"comment":"The real-world demonstration does not use range-only sensors: the 'pseudo range distances' are inferred indirectly from AirSLAM global point clouds produced by camera-based SLAM instances for all agents. These distances are deterministic functions of the very 3D positions that a range-only estimator is meant to recover, making the experiment unable to validate the claim of 'range-only measurements' or the estimation of a non-cooperative target. The statement that camera-derived ranges 'closely resemble real-world measurement uncertainty' is not quantified or compared with actual UWB or acoustic ranging statistics. The paper's headline claim of the 'first to demonstrate autonomous drone-to-drone interception ... with range-only measurements' is therefore not supported by the presented evidence.","section":"Section IV-B"},{"comment":"The encirclement error bound in Appendix B, ε2,2 = (3(α−1)²ϱ1 + 3(ϱ2+ˇv2)²)/(1−3α²), depends on the velocity compensation error bound ϱ2. Because the compensation in Eq. (11c) is a single impulse, ϱ2 cannot be taken as small for k>km; the bound must use ϱ2 = ˇv2. Consequently, the claimed ultimate encirclement error for Target 2 is not established under continuous target motion. The proof needs to be reworked with a realistic bound on ev,2.","section":"Appendix B"}],"minor_comments":[{"comment":"The term δ{k=km} is called a Dirac delta, but in discrete time it is a Kronecker delta; please correct the terminology.","section":"Eq. (11c)"},{"comment":"Please provide quantitative statistics of the pseudo-range noise (e.g., standard deviation, correlation with the SLAM estimates) to support the claim that they 'closely resemble real-world measurement uncertainty.'","section":"Section IV-B"},{"comment":"The claim of being the 'first to demonstrate' should be tempered or supported by a direct comparison with the closest prior range-only interception work, especially because the experiment uses pseudo ranges rather than actual range sensors.","section":"Section I"},{"comment":"The formula for r3, r3(−) − (r2−¯r)(v̂2(Ω3(0))+ι2)/(z2−z1), mixes velocity and distance terms and is dimensionally unclear; please clarify the intended time dependence.","section":"Section III-C, Case 3"},{"comment":"The constants ε1,1 and ε2,1 are defined with 'a' in place of 'α'; please make the notation consistent.","section":"Appendix B"},{"comment":"The experiment relies on offboard NUC processing and ground-based vision, so the phrase 'minimal onboard sensors' should be clarified to reflect that the processing is not onboard.","section":"Section IV-B"}],"recommendation":"major_revision","confidential_remarks":"The paper depends substantially on the authors' previous results (Lemmas from [30], [40], [43]); the current manuscript should either reproduce or clearly cite these results in a self-contained way. Also, the experimental section should be reframed as a system integration demo with visual SLAM, not as validation of range-only sensing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nRead this one with the experiment section open next to the estimator equations. The real news: the authors have assembled a 3D AS encirclement controller with vertical jitter and an adaptive zone-based transition from protecting a friendly target to encircling and then ramming a hostile one. That integrated system, with the first hardware demo attached, goes beyond their earlier 2D AS encirclement papers [34], [35]. The estimation law (RLS with PE excitation from the encirclement motion) and the Lyapunov-style convergence proofs are derived, not curve-fit, and the simulation shows position error around 5 cm once the drones switch to the hostile target. Credit where due: the analysis is careful, the assumptions are stated (bounded target velocity, PE shape), and the simulation parameters are design choices rather than fitted numbers.\n\nThe soft spots are real, and one is load-bearing. The hardware experiment feeds 'pseudo range distances' inferred from AirSLAM global point clouds into the estimator. Those clouds already come from camera-based SLAM that gives each drone's global position, so the pipeline effectively has the hostile drone's 3D position available. The claim that camera-derived ranges 'closely resemble real-world measurement uncertainty' is asserted, not shown; acoustic or UWB ranging has different noise, outages, and update rates. So the demonstration as reported does not validate the range-only headline. The authors are transparent about this in Section IV-B, but the conclusion overstates what the demo proves.\n\nSecond, Eq. (11c) has a Kronecker delta: the velocity compensator v̂₂ acts only at the single step k = kₘ and is zero afterwards. The proof of Theorem 1 in Appendix A assumes a bounded velocity error eᵥ, ₂ for all time, but with a one-shot compensator, eᵥ, ₂ is just the full unknown velocity for all subsequent steps. That leaves the convergence guarantee valid only in the trivial case of a stationary or near-stationary target. For a maneuvering hostile drone, the bound in Theorem 1 does not follow. This is fixable, but as written it is a gap.\n\nMinor issues: zone radii z₁, z₂, z₃ are never given numeric values in the paper; the real-world run is about 50 seconds with no repeated trials; and the system depends heavily on the authors' own prior results for Lemma 1, which is acceptable but keeps the novelty incremental.\n\nBottom line: this deserves a serious referee, but the next version needs either real range sensors on the drones or a much more careful statement about what the pseudo-range demo can support. The controller and estimator are worth engaging with; the empirical claim needs to catch up.","headline":"A mostly sound 3D range-only encirclement controller with a hardware demo that doesn't actually use real range sensors.","tokens_in":12929,"tokens_out":2722,"would_cite":false,"duration_ms":28106,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93C85"],"pacs":[],"model":"deepseek-v4-flash","headline":"Two guardian drones can encircle and intercept a hostile UAV using only noisy range measurements.","keywords":["range-only localization","target encirclement","anti-synchronization control","UAV interception","non-cooperative target","persistently exciting","drone defense","3D encirclement"],"falsifier":"Run the same estimator and controller on hardware where distances come from a real non-cooperative range sensor, such as a microphone array or UWB, with measured noise, and compare the achieved position and encirclement errors with the simulation's roughly 5 cm values under the same target trajectory. If the errors do not stay within the theoretical bound ϱ1 as the target maneuvers near the assumed maximum speed, or if the drones cannot hold the anti-synchronized geometry, the central claim fails.","tokens_in":11887,"feed_emoji":"🛸","tokens_out":8283,"duration_ms":85572,"temperature":0.7,"pith_summary":"Commercial drones carrying hazardous payloads are hard to stop with ground-based radar or vision, especially when GPS is jammed or the target is out of line of sight. This paper argues that two lightweight guardian drones, each measuring only noisy distances to a non-cooperative target, can estimate its 3D position, encircle it, and, if the target breaches a warning threshold, intercept it by collision. The authors propose a target position estimator that combines the two distance measurements with a recursive-least-squares update and a velocity compensator, plus an anti-synchronization controller that places the two drones on opposite sides of the target. They prove bounded estimation and encirclement errors under a persistently exciting motion condition, and they report a real-world demonstration in which two guardians switch from protecting a surface convoy to encircling and taking down a hostile drone. If correct, the approach offers an anti-drone capability that needs no surface guidance, no camera-based bearing tracking of the target, and no radio-frequency jamming.","feed_headline":"Two drones, only noisy distances, encircle and intercept a hostile UAV","feed_subtitle":"If right, lightweight guardian drones can defend convoys without GPS, radar, or surface guidance.","key_machinery":"The central object is the intersection circle of the two range spheres centered at the drones: with two noisy distances d1 and d2 and the known inter-drone distance d12, the target lies on a circle with computable center c and radius ς. The estimator uses the displacement c − c(−) as a velocity compensator for the target's unknown motion, and feeds a recursive-least-squares update with forgetting factor γ1 on the observation ϖ = q12^T q2, so that persistently exciting relative motion q12 drives the position error to a bound. The controller is built on the anti-synchronization shape vector ζ(r,ν,k) = r(sin(νkπ), cos(νkπ), g(k))^T, whose vertical jitter g(k) keeps the sequence persistently exciting; Drone 1 is driven toward +ζ and Drone 2 toward −ζ relative to the target, giving maximum geometric coverage and a unique target position from two ranges.","core_discovery":"The paper claims that range-only measurements from two moving drones are sufficient to track, encircle, and intercept a non-cooperative 3D target, provided the drones' motion keeps the relative geometry persistently exciting. The target position estimator uses, at each step, the circle formed by the intersection of two range spheres; the motion of that circle's center is used to compensate for the target's unknown velocity. A recursive least-squares estimator with exponential forgetting factor, driven by the projection of the estimated position onto the inter-drone baseline, yields an estimation error bounded by ϱ1 once the forgetting factor γ1 ≤ 1/2. The anti-drone controller then steers Drone 1 to +ζ and Drone 2 to −ζ relative to the target—anti-synchronized positions on opposite sides—with the radius shrinking as the hostile target enters warning and take-down zones; Theorem 2 gives bounded encirclement error for controller gain α in (−1/√3, 1/√3]. The authors validate the scheme in numerical simulation and in a 50-second real flight where two low-cost drones, using distances inferred from camera-based SLAM point clouds, first orbit a protected surface vehicle and then encircle and collide with a hostile drone.","pith_inferences":["The experimental component uses pseudo-ranges extracted from camera-based SLAM point clouds rather than true acoustic or radio range sensors; a natural extension would be to replay the same controller with real microphone-array or UWB ranging hardware, whose noise statistics differ, to see whether the roughly 5 cm simulation error persists.","Because the velocity compensator improves as the drones approach the target, the method seems best suited to terminal-phase interception; long-range detection would still need another sensor to bring the drones into the warning zone.","The anti-synchronized geometry suggests a testable scaling: adding more guardian drones in anti-synchronized pairs should reduce the per-drone motion needed for persistent excitation, which could be checked by simulating three or more drones with the same estimator.","A hostile target that maneuvers to break the persistently exciting condition—for instance by matching the drones' circular motion—could in principle drive the estimator toward its error bound; the paper's guarantees do not cover targets that are actively optimizing against the estimator."],"forward_implications":["If the estimation and control claims hold, two small drones with only ranging hardware can protect a convoy or ship without any ground-based guidance station.","The same controller can transition seamlessly from a protective orbit around a friendly target to a monitoring and encirclement orbit around a hostile drone, with the orbit radius shrinking as the threat approaches.","Range-only sensing removes the need for a camera-based bearing pipeline, making the scheme robust in darkness, fog, GPS denial, and radio jamming.","Under the persistently exciting encirclement motion, the theoretical bounds guarantee that position estimation error and encirclement error remain finite, and the error shrinks as the drones get closer to the target.","When the hostile target enters the take-down zone, the controller can reduce the radius to a collision, meaning a kinetic kill by the guardian drone itself."],"supporting_citations":[{"why":"Supplies the persistently exciting adaptive relative localization framework and the recursive-least-squares estimator structure that the target position estimator adapts.","marker":"[30]"},{"why":"Establishes anti-synchronization encirclement control for multiple non-cooperative targets, forming the basis of the proposed anti-drone controller.","marker":"[34]"},{"why":"Previous range-only anti-synchronization encirclement result that the paper extends from planar motion to 3D interception.","marker":"[35]"},{"why":"Supports the claim that observing the target from opposite sides maximizes coverage and positioning accuracy.","marker":"[39]"},{"why":"Provides the visual SLAM point clouds from which the experiments infer pseudo-range distances between drones and targets.","marker":"[41]"},{"why":"Supplies the exponential convergence result for recursive least squares with forgetting factor that the proof of Theorem 1 relies on.","marker":"[43]"},{"why":"Demonstrates microphone-array detection of nearby UAVs, the intended physical ranging modality for non-cooperative targets.","marker":"[38]"}],"fun_headline_variants":["Range-only drone duo encircles and kills hostile UAVs autonomously","Autonomous drones need only distance to encircle and intercept UAVs","No GPS or radar: two range-only drones track and ram hostile UAV","Two UAVs, noisy ranges only: autonomous 3D intercept of hostile drone","Guardian drones shift from protect to attack with only range data"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The demonstration assumes that pseudo-range distances obtained from camera-based SLAM point clouds behave like true noisy range-only measurements; if real acoustic or radio ranging is much noisier or missing, the experimental validation does not support the range-only claim.","fun_headline_variants_meta":{"raw":{"variants":["Range-only drone duo encircles and kills hostile UAVs autonomously","Autonomous drones need only distance to encircle and intercept UAVs","No GPS or radar: two range-only drones track and ram hostile UAV","Two UAVs, noisy ranges only: autonomous 3D intercept of hostile drone","Guardian drones shift from protect to attack with only range data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001332,"raw_usage":{"total_tokens":5453,"prompt_tokens":1017,"completion_tokens":4436,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":4342}},"tokens_in":633,"tokens_out":4436,"duration_ms":32644,"temperature":1.0,"reasoning_tokens":4342,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:36:40.095414+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same estimator and controller on hardware where distances come from a real non-cooperative range sensor, such as a microphone array or UWB, with measured noise, and compare the achieved position and encirclement errors with the simulation's roughly 5 cm values under the same target trajectory. If the errors do not stay within the theoretical bound ϱ1 as the target maneuvers near the assumed maximum speed, or if the drones cannot hold the anti-synchronized geometry, the central claim fails.","supporting_citations":[{"cited_title":"Persistently excited adaptive relative localization and time-varying formation of robot swarms,","cited_arxiv_id":null,"evidence_quote":"Supplies the persistently exciting adaptive relative localization framework and the recursive-least-squares estimator structure that the target position estimator adapts."},{"cited_title":"Multiple non- cooperative targets encirclement by relative distance based positioning and neural anti-synchronization control,","cited_arxiv_id":null,"evidence_quote":"Establishes anti-synchronization encirclement control for multiple non-cooperative targets, forming the basis of the proposed anti-drone controller."},{"cited_title":"Non-cooperative stochastic target encirclement by anti-synchronization control via range-only measurement,","cited_arxiv_id":null,"evidence_quote":"Previous range-only anti-synchronization encirclement result that the paper extends from planar motion to 3D interception."},{"cited_title":"Optimal tdoa sensor-pair placement with uncertainty in source location,","cited_arxiv_id":null,"evidence_quote":"Supports the claim that observing the target from opposite sides maximizes coverage and positioning accuracy."},{"cited_title":"Airslam: An efficient and illumination-robust point-line visual slam system,","cited_arxiv_id":null,"evidence_quote":"Provides the visual SLAM point clouds from which the experiments infer pseudo-range distances between drones and targets."},{"cited_title":"Exponential convergence of recursive least squares with exponential forgetting factor,","cited_arxiv_id":null,"evidence_quote":"Supplies the exponential convergence result for recursive least squares with forgetting factor that the proof of Theorem 1 relies on."},{"cited_title":"Detection of nearby uavs using a multi-microphone array on board a uav,","cited_arxiv_id":null,"evidence_quote":"Demonstrates microphone-array detection of nearby UAVs, the intended physical ranging modality for non-cooperative targets."}],"review_version":1}