{"id":"c8e481e8-e960-47cd-a48a-2fc095590d5e","arxiv_id":"2505.16087","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A decentralized robot swarm controller that switches formations when a narrow passage is detected, with a Lyapunov stability proof and simulation/SIL evidence.","lead":"This paper proposes an event-based controller that lets a swarm of robots switch from a normal formation into a single-file line to pass through narrow passages, then reform. The authors test it in simulation and software-in-the-loop flights, and report that it beats two earlier potential-field methods in success rate, speed, travel time, and energy use.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1 does not follow from the proof: tailgating mode converges to a line, not the task configuration, and the mode-wise Lyapunov analysis omits migration, avoidance, switching, and the acceleration-level implementation.","rationale":"The reader's weakest assumption about the fully connected graph is real and relevant: the Laplacian spectrum in Eq. (21) indeed fails under limited communication range, and the paper's decentralized claim is weakened by that assumption. However, the more load-bearing problem is that Theorem 1's conclusion does not match the controller's intended behavior: tailgating mode is explicitly designed to produce a line configuration, and the proof analyzes each behavior in isolation without accounting for the always-active migration and avoidance terms, the event-triggered mode switches, or the acceleration-level implementation. These omissions mean the theorem is not a valid guarantee for the system in Eq. (14). This reinforces the reader's conditional verdict rather than overturning it: the simulations, public code, and SIL tests provide empirical evidence that the controller can work in the tested scenarios, but the theoretical stability claim needs substantial revision or qualification. I therefore agree partially with the reader's identified weakness and recommend keeping the verdict unchanged.","tokens_in":12215,"tokens_out":4128,"duration_ms":38408,"concrete_test":"Run the public code in an environment containing a long, straight corridor (e.g., 50 m long, 0.8 m wide) so that tailgating mode persists through the entire traversal, and record the steady-state inter-robot offsets after the swarm exits. If the final configuration is a straight line rather than the assigned V-shape or polygon, Theorem 1's claim of asymptotic convergence to the task configuration is contradicted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 1: under control law (14), the TVF 'asymptotically converges to the desired task configuration.' The proof in Section 3.3 does not establish this. First, in tailgating mode control law (8)/(14) deliberately drives each robot to align behind a leader at distance d_ref; the proof's own V_t analysis concludes only a straight-line configuration. Thus the invariant set of tailgating mode is the safe line configuration, not the task configuration. If a robot remains in tailgating mode, e.g., in a long corridor, convergence to the task configuration is false, so Theorem 1 overclaims. Second, even the formation-mode analysis uses ˙P = -k_f H P + B (Eq. 23) with a static κ, a fully connected graph, and no migration or avoidance terms. In the actual controller, κ is piecewise constant, mode switches occur via Algorithm 1, migration (9) and avoidance (10)-(13) are always active, and the commanded signal is acceleration after differentiation and saturation (17)-(19). Each Lyapunov derivative is computed using only a subset of the control terms, and separate negativities in (36) do not constitute a LaSalle argument for the coupled switched system. The theorem is therefore unsubstantiated for the controller actually proposed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes an event-based reconfiguration controller (ERC) for a swarm of robots navigating narrow spaces. Each robot runs a local algorithm that selects between a formation mode and a tailgating mode on the basis of sensed corridor width, using artificial-potential-field behaviors for migration, formation maintenance, tailgating, inter-agent avoidance, and obstacle avoidance. The authors prove, in Theorem 1, that the time-varying formation asymptotically converges to the desired task configuration under control law (14). They report simulations in forest-like and cave-like environments, comparisons with APF and IAPF baselines, and software-in-the-loop tests with three Hummingbird UAVs. The central theoretical claim, however, is not established by the proof as written, and the paper's simulation and SIL results should be understood as empirical demonstrations rather than as consequences of the stated theorem.","tokens_in":12521,"tokens_out":4862,"duration_ms":46340,"significance":"If the theorem were correct, the paper would offer a distributed, event-triggered alternative to centralized reconfiguration planning, with a relatively simple potential-field controller and a public implementation. The empirical comparison is useful: the ERC achieves a higher success rate than the APF/IAPF baselines in the tested scenarios, and the code is publicly available. The weakness is that the main advertised guarantee, the asymptotic convergence to the task configuration, does not follow from the supplied Lyapunov analysis; the proof analyzes individual behaviors in isolation and does not handle the actual switched, acceleration-level, locally-sensed system. Because the theoretical contribution is load-bearing for the paper's framing, the manuscript needs substantial revision of the stability analysis or a correspondingly weakened theorem before it can be accepted.","major_comments":[{"comment":"Theorem 1 claims that under control law (14) the TVF asymptotically converges to the desired task configuration. The proof's own tailgating analysis, Eqs. (26)-(28), concludes only that robot i aligns behind its leader at distance d_ref in a straight-line configuration. Thus the invariant set of tailgating mode is the safe line configuration, not the desired task shape; in a long corridor where robots remain in tailgating mode, Theorem 1 is false as stated.","section":"§3.3, Theorem 1"},{"comment":"The proof analyzes a first-order, velocity-level closed loop ˙P = −k_f H P + B with a constant κ and no switching, whereas the actual controller is executed at acceleration level (Eqs. (17)-(19)), includes the always-active migration and avoidance terms (9)-(13), and has piecewise-constant, robot-dependent κ from Algorithm 1. The separate negativities of ˙V_F and ˙V_T in (36) do not constitute a Lyapunov or LaSalle argument for the coupled switched system with mode transitions driven by local sensing. The theorem is therefore unsubstantiated for the controller actually proposed.","section":"§3.3, Eqs. (20)-(25) and (35)-(36)"},{"comment":"The proof requires the sensing/communication graph G to be fully connected ('As G is fully connected') so that the Laplacian has the stated spectrum and HB=nB holds. This contradicts the decentralized description in Remark 2 and Section 2.1, where each robot uses only local sensors and peer communications; for a general graph the spectrum is not as claimed, and the convergence argument does not carry over to larger swarms or limited communication ranges.","section":"§3.3, Eq. (21)"},{"comment":"The collision-avoidance Lyapunov computation is incomplete. The derivative ˙V_i in (31) contains (v_i − v_j), but the substitution into (33) uses only robot i's avoidance contribution and ignores the leader/tailgating, migration, and avoidance control of robot j and of all other active terms. Consequently the claimed negativity ˙V_i < 0 is not established for the full coupled system with control law (14).","section":"§3.3, Eqs. (31)-(34)"},{"comment":"The implementation uses a per-robot scaling factor κ, as Fig. 8 explicitly states, while Eq. (6) is written with a single κ multiplying the relative offsets. If each robot uses its own κ_i, the desired offsets are not those of a common scaled formation, and the formation-error term in (20) needs to be re-examined. The manuscript should clarify how a shared formation shape is encoded when the mode decision and κ are evaluated locally.","section":"§3.2 and §4.2, Eq. (6) and Fig. 8"}],"minor_comments":[{"comment":"In the comparison paragraph, 'AFP' appears to be a typo for 'APF'.","section":"§4.3"},{"comment":"The formation condition (3) is written as a sum of norms tending to zero; this is mathematically equivalent to requiring all robots to have the same offset p_i − δ_i, but the notation should be clarified so that readers do not mistake it for a statement forcing all δ_i equal.","section":"§2.2, Definition 2"},{"comment":"The pseudocode returns a desired velocity but the dynamics (1) require acceleration; the connection to Eq. (17) should be stated more explicitly in the algorithm itself, since the proof in Section 3.3 operates at the velocity level.","section":"§3.2, Algorithm 1"},{"comment":"The SIL test is reported with three UAVs only; a sentence explaining why this small scale is representative of the 5-robot simulation results would improve the validation discussion.","section":"§4.4"}],"recommendation":"major_revision","confidential_remarks":"The paper's empirical results and open-source implementation are valuable, but the theoretical claim is not currently supported. If the authors can rewrite the stability analysis for the actual switched, acceleration-level system, or substantially weaken Theorem 1 to a statement about the formation mode alone and present the narrow-passage behavior as an empirically validated heuristic, the manuscript could become publishable. As it stands, the proof does not match the controller it is supposed to verify."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — read this if you work on behavior-based swarm navigation. It is a decent engineering paper with public code and a working idea, not a deep theory paper. What is new: an event-triggered switch between formation and tailgating modes, with a local width estimator used to scale the formation (Eqs. 15–16), plus five APF behaviors combined into two modes. The comparison against APF and IAPF shows better success rates and travel times, and the software-in-the-loop test with a quadrotor model adds credibility. The source code is public, which is real evidence.\n\nThe soft spots matter. Theorem 1 says the TVF converges to the desired task configuration under (14), but the proof doesn't show that. In tailgating mode, the V_t argument only shows convergence to a straight line with spacing d_ref, as the paper itself acknowledges; that is the safe configuration, not the task configuration. So if a robot stays in tailgating mode, the claimed convergence to the task shape is false. Also, each Lyapunov derivative is computed on a subset of control terms, with a static κ, a fully connected graph, and no migration, avoidance, or switching in the model. The actual controller has piecewise-constant κ, mode switches, always-active migration and avoidance, and acceleration-level commands after differentiation and saturation. The mode-wise sum of Lyapunov functions in (35)–(36) is not a LaSalle argument for the coupled switched system. So the theorem is unsubstantiated for the controller actually proposed.\n\nThe comparison is also only 10 runs per method, with no variance or statistical test, and the gains are hand-tuned. The novelty over the authors' own V-shape controller [16] is not quantified. These are fixable but real gaps.\n\nI agree with the reader's conditional verdict. The core event-based reconfiguration idea is plausible and independently testable; the proof needs a major rewrite or a reduction in claim. I would send it to peer review, because the engineering contribution and the public code justify referee time. I would not cite the theorem in its current form.","headline":"A practical, code-backed two-mode reconfiguration controller for swarm navigation, but Theorem 1 overclaims and the Lyapunov proof doesn't cover the implemented controller.","tokens_in":13039,"tokens_out":1827,"would_cite":false,"duration_ms":15734,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proposes an event-based reconfiguration controller that switches a robot swarm between its task formation and a single-file tailgating configuration to pass narrow spaces, and proves the switched system converges asymptotically.","keywords":["multi-robot system","time-varying formation","reconfiguration control","swarm robotics","event-triggered control","artificial potential field","narrow space navigation","Lyapunov stability"],"falsifier":"Run control law (14) with robots whose communication range is too short to keep the graph fully connected, and record whether the formation still converges to the task configuration; a divergence in that condition would falsify the claim that the TVF converges as stated, because the theorem's proof explicitly depends on the fully connected Laplacian.","tokens_in":12041,"feed_emoji":"🤖","tokens_out":8115,"duration_ms":61508,"temperature":0.7,"pith_summary":"This paper proposes an event-based reconfiguration controller that lets a robot swarm change its formation while moving through narrow passages such as valleys, tunnels, and corridors. The controller has two modes: a formation mode that holds the desired task shape, and a tailgating mode in which robots line up single-file to pass through gaps too narrow for the full shape. Each robot decides which mode to use from its own local sensor readings, so the decision is distributed rather than centralized. The paper proves with Lyapunov functions that, under the proposed control law, the time-varying formation converges asymptotically to the desired task configuration. Simulation comparisons and software-in-the-loop tests are reported as supporting evidence.","feed_headline":"Swarm reconfigures to slip through narrow gaps","feed_subtitle":"An event-based controller switches robot formations between task shape and single-file tailgating, with a stability proof.","key_machinery":"The load-bearing object is the event-triggering mode switch around control law (14), which sums behavior velocities and selects between the formation term $v^f_i$ and the tailgating term $v^t_i$ depending on the detected environment width. The formation term uses a scaled task configuration $\\kappa\\delta^*$, and the tailgating term makes each robot follow the nearest robot in front at a desired distance $d_{\\mathrm{ref}}$. The event trigger computes the environment width $w_e$ from the left and right nearest obstacle points and compares it with the threshold $\\lambda r$. Stability is carried by Lyapunov functions for each behavior, with the formation-mode proof relying on the Laplacian matrix of the sensing graph $G$ and the identity $H B = n B$ for the bias vector.","core_discovery":"The paper's central claim is that a single distributed control law can make a robot swarm reconfigure from a task formation into a safe single-file configuration and back, so that the formation survives narrow passages. The law combines five potential-field behaviors: formation, tailgating, migration, inter-agent avoidance, and obstacle avoidance. An event-triggering rule uses each robot's local range data to estimate the width of the environment and the width of the current formation; when the passage width drops below a threshold, the robot switches to tailgating, and the scaling factor $\\kappa$ contracts the formation. Theorem 1 states that under control law (14) the time-varying formation described by (1) asymptotically converges to the desired task configuration. The proof builds separate Lyapunov functions for the formation, tailgating, and collision-avoidance behaviors and combines them mode by mode.","pith_inferences":["The controller's reliance on a fully connected graph means the published theorem is really a guarantee for small or densely communicating swarms; extending the Lyapunov argument to switching or disconnected topologies would be the natural next step and would make the stability claim match the paper's stated decentralization.","The same event-triggered shrinking-and-single-file mechanism could transfer to ground vehicles or heterogeneous robots, since the controller only needs local range data and a rule for picking a front leader; the paper does not test this.","The threshold $\\lambda r$ and the desired tailgating distance $d_{\\mathrm{ref}}$ together define an implicit safety-speed trade-off; sweeping them across passage widths could produce a design curve for choosing how early a swarm abandons its task shape, which the paper does not explore."],"forward_implications":["A swarm using the ERC can pass through a corridor without a central planner, with each robot choosing formation or tailgating mode from its own sensor data.","The same control law covers nominal formation flight and the single-file safe configuration, so no separate switching controller is needed.","The formation contracts by the scaling factor $\\kappa = (w_e - 2r)/w_f$ when the passage is wider than the safety margin but narrower than the task shape, so reconfiguration happens continuously rather than only at the trigger threshold.","If the convergence theorem holds, the swarm's shape is restored after the passage, so the task configuration is not lost during reconfiguration."],"supporting_citations":[{"why":"Supplies the time-varying formation definition and the double-integrator robot model that Theorem 1 is stated for.","marker":"[6]"},{"why":"Establishes the behavior-based optimized flocking approach in confined environments that the proposed controller extends and later compares against.","marker":"[9]"},{"why":"Prior self-reconfigurable V-shape formation work in narrow spaces that motivates the reconfiguration problem addressed here.","marker":"[16]"},{"why":"Provides the improved artificial potential field baseline and the inter-agent and obstacle avoidance force formulas used in equations (12)-(13).","marker":"[33]"},{"why":"Supplies the LaSalle invariance principle used in the formation-mode stability proof.","marker":"[34]"},{"why":"Provides the quadrotor simulation environment used for the software-in-the-loop validation tests.","marker":"[35]"}],"fun_headline_variants":["Event-triggered swarm reconfigures to fit narrow passages","Swarm shrinks to single-file via event-based reconfiguration","Robot swarm mode-switches to slip through tight gaps","Event-based control proves swarm can navigate corridors"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes the sensing and communication graph is fully connected, so every robot effectively sees every other robot; in a larger or more sparse swarm that assumption can fail and the Lyapunov argument no longer applies.","fun_headline_variants_meta":{"raw":{"variants":["Event-triggered swarm reconfigures to fit narrow passages","Swarm shrinks to single-file via event-based reconfiguration","Robot swarm mode-switches to slip through tight gaps","Event-based control proves swarm can navigate corridors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000202,"raw_usage":{"total_tokens":1367,"prompt_tokens":913,"completion_tokens":454,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":529,"completion_tokens_details":{"reasoning_tokens":388}},"tokens_in":529,"tokens_out":454,"duration_ms":4856,"temperature":1.0,"reasoning_tokens":388,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T15:06:33.578081+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run control law (14) with robots whose communication range is too short to keep the graph fully connected, and record whether the formation still converges to the task configuration; a divergence in that condition would falsify the claim that the TVF converges as stated, because the theorem's proof explicitly depends on the fully connected Laplacian.","supporting_citations":[{"cited_title":"Time- varying formation control for unmanned aerial vehi- cles: Theories and applications,","cited_arxiv_id":null,"evidence_quote":"Supplies the time-varying formation definition and the double-integrator robot model that Theorem 1 is stated for."},{"cited_title":"Optimized flocking of autonomous drones in confined environments,","cited_arxiv_id":null,"evidence_quote":"Establishes the behavior-based optimized flocking approach in confined environments that the proposed controller extends and later compares against."},{"cited_title":"Self-reconfigurable V-shape formation of multi- ple UA Vs in narrow space environments,","cited_arxiv_id":null,"evidence_quote":"Prior self-reconfigurable V-shape formation work in narrow spaces that motivates the reconfiguration problem addressed here."},{"cited_title":"Multi uav cluster control method based on virtual core in improved artificial potential field,","cited_arxiv_id":null,"evidence_quote":"Provides the improved artificial potential field baseline and the inter-agent and obstacle avoidance force formulas used in equations (12)-(13)."},{"cited_title":"Invari- ance principles and observability in switched sys- tems with an application in consensus,","cited_arxiv_id":null,"evidence_quote":"Supplies the LaSalle invariance principle used in the formation-mode stability proof."},{"cited_title":"Furrer, M","cited_arxiv_id":null,"evidence_quote":"Provides the quadrotor simulation environment used for the software-in-the-loop validation tests."}],"review_version":1}