{"id":"c159af35-00de-40bb-ae49-357f518668a8","arxiv_id":"2607.10170","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A distributed CBF-QP controller with a hierarchical 'splay' geometry acquires rigid communication graphs from non-rigid initial graphs under limited sensing while guaranteeing collision avoidance.","lead":"This paper presents a distributed controller that lets a team of robots with limited sensing ranges build a rigid communication network from a non-rigid start while avoiding collisions. It matters because rigid networks are what make multi-robot formations hold their shape, and most prior work assumes such a network already exists.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4 does not prove feasibility of the multi-constraint C3BF-QP; two opposing collision-cone constraints can be jointly infeasible, so collision-free rigidity acquisition is not guaranteed.","rationale":"The reader's weakest_assumption is Assumption 4. I agree that Assumption 4 is a separate, unvalidated premise and that the recursive bounds carry global information; the paper even concedes leaders use global frame information. But I think the most load-bearing point for the central claim is the missing feasibility proof of the QP in Theorem 4. Even if Assumption 4 is granted, Theorem 4's safety conclusion does not follow from Theorem 2 unless a single \\hat q_i can satisfy all active constraints. The paper provides no such argument. Indeed the conservative bounds in Assumption 4 enter with signs that shrink the feasible set, so the concern is not an artifact of implementation but a potential counterexample to the formal guarantee. A concrete three-agent scenario with opposing collision cones would settle it. If infeasibility is exhibited, the theorem's claim of guaranteed safety is false; if the authors can prove non-emptiness (e.g., via a common Lyapunov/feasibility argument or by adding a feasibility assumption), the paper's conditional acceptance remains appropriate. Thus I do not change the overall verdict but flag this as the key issue to resolve.","tokens_in":158,"tokens_out":10473,"duration_ms":145820,"concrete_test":"Check feasibility of (14) at a reachable state with two active constraints: set i at origin, j1 at d e_x moving with -v e_x, j2 at -d e_x moving with +v e_x, dynamics per (1), parameters satisfying Assumptions 1–4 and Lemma 2/3. Compute c_j = (L'_fij+κ(b_j)) / ((Γig(Bp+Bv))^{-1} Γ_i ||ξ_ij||^2). If c_1 + c_2 < 0 while ξ_i1 ≈ -ξ_i2, the QP is infeasible, disproving the assertion that safety follows from Theorem 2. This can be done analytically (linear-programming feasibility in R^2) or in a 3-agent simulation by logging the QP solver status at the first instant both constraints are active.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central safety guarantee in Theorem 4 (Eq. 14) is asserted by saying forward invariance 'follows directly from Theorem 2,' but Theorem 2 only certifies a single pair (i,j) with control u_i = g_i ξ_ij ξ_ij^T \\hat u_i. In the multi-obstacle QP, one \\hat q_i must simultaneously satisfy, for every j ∈ \\tilde N_i(t), L'_fij b + L'_hij b \\hat q_i + κ(b) ≥ 0 and ξ_ij^T \\hat q_i ≤ 0. These are linear inequalities whose normals are the vectors ξ_ij; no proof is given that the intersection is nonempty along the closed-loop trajectory. Feasibility is made harder by the conservative bounds in Assumption 4: L'_hij b \\hat q_i = -(Γig(Bp+Bv))^{-1} Γ_i ||ξ_ij||^2 ξ_ij^T \\hat q_i and L'_fij contains negative (Bp+Bv)-dependent terms. If L'_fij+κ(b)<0, the constraint forces ξ_ij^T \\hat q_i to be ≤ a negative number scaled by 1/(Bp+Bv). Two sensed agents approaching from opposite sides give roughly opposite normals, and the two constraints cannot be met simultaneously (the feasible set is empty). Thus the advertised 'guaranteed collision-free motion' is not established at exactly the point where multiple safety constraints interact. This is an internal proof gap, not merely an implementation detail.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies distributed rigid communication-graph acquisition and collision avoidance for n planar heterogeneous nonlinear agents (1) with heterogeneous limited sensing ranges. It assumes a time-invariant leader-follower iLDAG spanning forest (Def. 2) and proposes a 'splay scheme' (Def. 6) under geometric orbit decay (Def. 7), a hierarchical second-order consensus controller (Thm. 3), and a C3BF-QP safety filter (Thms. 2 and 4). It claims: (i) every follower acquires a cross-family maintenance link (Lemmas 1 and 3), making the maintenance graph rigid in the sense of Laman's theorem; (ii) inter-agent safety is preserved despite accelerating nonlinear agents; (iii) followers need no global positions. The claims are supported by appendix proofs and by a 27-agent simulation and a 5-robot hardware experiment with 9 virtual agents.","tokens_in":21192,"tokens_out":11240,"duration_ms":131926,"significance":"If the technical claims were fully established, the paper would contribute a constructive, distributed approach to an under-studied problem: acquiring (rather than maintaining) rigid sensing/communication graphs under range limits and collision constraints, with nonlinear heterogeneous dynamics and only relative local measurements for followers. The iLDAG/splay construction is original and explicit, the edge-addition/rigidity argument via Lemma 1 is plausible, and the authors provide complete appendix derivations, parameter inequalities (Lemmas 2-3), and experimental validation. The paper also clearly identifies the gap relative to rigidity-maintenance literature. However, the main safety theorem is not presently established because the multi-constraint QP feasibility and rank-one projection consistency are not proved; this limits confidence in the central guarantee.","major_comments":[{"comment":"The safety guarantee of Theorem 4 is not established. Theorem 2 certifies only a single obstacle pair (i,j) with control u_i = g_i ξ_ij ξ_ij^T \\hat u_i. In Theorem 4, one \\hat q_i must satisfy all constraints j∈\\tilde N_i(t). For each j the CBF constraint is L'_fij b + L'_hij b \\hat q_i + κ(b) ≥ 0. Since L'_hij b \\hat q_i = -A_j ξ_ij^T \\hat q_i with A_j > 0, near the boundary this enforces ξ_ij^T \\hat q_i ≤ c_j with c_j < 0. Two sensed agents approaching from opposite directions give opposite ξ_ij, making the feasible set empty. The proof in Appendix .1 ('feasible because Γ...≠0') only addresses one inequality, and no feasibility argument is given for the intersection of half-spaces. Moreover, the same theorem uses a single projection direction ξ_ij while constraints involve all j; for a neighbor m ≠ j, the lower-bound derivation used in Theorem 2 does not apply to u_i unless the project","section":"§4.2, Theorem 4 (Eq. 14)"},{"comment":"The recursive accumulated bounds Bp and Bv are load-bearing: they enter every CBF lower bound in Theorem 2 and the disturbance bound γ_ij in Theorem 3. Assumption 4 says only that the parent 'communicates' Bp(i,pr(i)) = ||p_ipr(i)|| + Bp(pr(i),pr^{(2)}(i)) (and similarly Bv). For the root, Bp(i,l)=||p_l||+||p_il||, so followers receive a scalar depending on the leaders' absolute position norm. This is difficult to reconcile with the advertised claim that followers use only relative positions/velocities and have no global information. More importantly, no protocol is given for updating Bp/Bv, while the proofs need the inequalities ||p_i|| ≤ Bp(i,pr(i)) to hold at all times. Delayed, noisy, or lost packets would break the bound and invalidate both the CBF lower bound and the consensus disturbance estimate. This assumption should be stated as a dynamic invariant with a communication/measure","section":"§2.7, Assumption 4"},{"comment":"Condition (13) is not a checkable design condition as stated. The variable γ* is used but never defined in the theorem statement; γ(t) is defined through γ_ij(t), which itself depends on the trajectory ||x_im(t)||, on Bp+Bv, and on the gains k_p,k_v. The proof's Remark 6 replaces it by an implicit equation R = C R_0 + C∆/(μ* - Cγ*), γ* = R'_0 + γ_1 R, making (13) an implicit region-of-attraction condition rather than an explicit inequality. Since Theorem 3 underpins convergence to the splay configuration and therefore the link-acquisition argument, the paper needs a constructive verification procedure or a clearly stated set of sufficient explicit inequalities.","section":"§4.1, Theorem 3 (Eq. 13)"},{"comment":"Even if QP feasibility were granted, the transition from asymptotic consensus (Theorem 3) to finite-time rigidity acquisition in Problem 1 is only sketched. The proof states that non-parent neighbors leave \\tilde N_i(t) and the 'nominal controller is thus recovered in finite time,' but consensus is asymptotic and T(ε) in Theorem 3 only gives an ultimate bound. To guarantee acquisition of the cross-family link by a finite t', the paper must quantify an ε small enough that the sensing-range inequalities in Lemma 3 remain satisfied despite the tracking error. The statement 'This solves Problem 1' does not follow from the displayed arguments.","section":"§4.2, proof of Theorem 4"}],"minor_comments":[{"comment":"The caption refers to 'Lemma 2, 3, 4', but there is no Lemma 4 in the manuscript.","section":"Figure 5 caption"},{"comment":"The phrase 'All the leaders form a rigid subgraph with at least 2n_l - 3 edges' followed by 'minimum edge count' is slightly confusing because the leader subgraph may have more than the minimum; the inequality should be stated explicitly as |E_leader| ≥ 2n_l - 3.","section":"Section 2.3, Definition 2"},{"comment":"The comparison with rigidity-maintenance methods is qualitative: a single target-enclosing scenario under one representative controller is shown to fail. This does not quantify performance against existing algorithms and should be labeled as an illustrative comparison, not a benchmark.","section":"Section 5.1"},{"comment":"Assumption 3 says the control input norm ||u_j|| can be measured or communicated, but Assumption 4 additionally requires Bp and Bv; the two assumptions should be consolidated or cross-referenced so that the full set of communicated quantities is explicit.","section":"Assumption 3"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the central theorem's feasibility gap is serious. If the authors can prove (or add a conservative feasibility condition for) the multi-obstacle C3BF-QP and clarify the rank-one projection, the paper could be acceptable. The reliance on recursive absolute-norm bounds should also be disclosed in the abstract/claims. I would not recommend publication before these points are resolved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, the core construction — an interleaved layered DAG spanning forest with a splay geometry that gives every follower a distinct target, so each acquires exactly one cross-family link within sensing range — is genuinely new and clever. The paper makes a good case that rigidity acquisition, not maintenance, is the problem worth solving, and that case holds up. Second, the central safety guarantee in Theorem 4 is not actually proven. The proof says forward invariance 'follows directly from Theorem 2,' but Theorem 2 certifies a single pair (i,j) with control restricted to the rank-one form u_i = g_i ξ_ij ξ_ij^T û_i. The multi-obstacle QP in (14) must satisfy one L'_f/L'_h constraint plus the half-space constraint ξ_ij^T q̂_i ≤ 0 for every sensed j, and nobody shows the feasible set is nonempty along the closed-loop trajectory. The stress-test concern is fair: two agents approaching from roughly opposite directions produce constraints ξ^T q̂ ≤ u1 and −ξ^T q̂ ≤ u2, which can be jointly infeasible, especially since the conservative Bp/Bv bounds push the u's negative. So the advertised 'guaranteed collision-free motion' has a real proof gap at the point where multiple safety constraints interact. The simulations and hardware runs suggest the approach works in practice, but the theorem overstates what is established.\n\nWhat is good: the iLDAG/splay idea is new relative to the cited literature; parameters are derived from inequalities rather than fit to a target; the rigidity argument via Laman conditions and Lemma 1 edge-addition is sound; and the leader-rigid-subgraph prerequisite is weaker than assuming the whole graph is rigid — Remark 2 acknowledges it. The self-citation to C3BF [14] is a legitimate prior method, not circularity.\n\nSoft spots, in proportion: (1) the multi-constraint feasibility gap above — this is the main issue; (2) Assumption 4 is load-bearing in every CBF constraint, and since Bp includes the leader's global position norm (for leader-parented agents), the 'no global absolute positions' claim is strained; (3) smaller items: Example 3's algebra does not parse — the quadratic 25α − 50α² − 2 > 0 does not follow from the stated inequality — and the desired splay position p*_ij depends on the parent's global orientation θ_j, so it is not obvious how a follower computes it. No code or data is released, and the comparison experiment is qualitative.\n\nBottom line: this is a serious paper for the multi-robot formation control audience. It deserves peer review — the referee should push for a feasibility argument (or an explicit feasibility assumption) in Theorem 4, a concrete implementation story for Bp/Bv that respects the sensing model, and a cleanup of the arithmetic. Expect major revision.","headline":"Genuinely new construction for rigidity acquisition from a non-rigid graph under limited sensing — but Theorem 4's safety guarantee has an unproven multi-constraint feasibility step, and Assumption 4 quietly weakens the 'no global positions' claim.","tokens_in":21732,"tokens_out":8978,"would_cite":true,"duration_ms":92466,"reading_group":"yes","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 claims that a distributed controller lets robots with limited, heterogeneous sensing ranges acquire a rigid communication graph from an initially non-rigid one, while guaranteeing inter-agent collision avoidance and requiring no","keywords":["rigid graph acquisition","communication graph rigidity","control barrier functions","inter-agent safety","limited sensing range","leader-follower consensus","splay formation","non-rigid communication graphs"],"falsifier":"Run the controller with a single follower whose parent deliberately reports Bp/Bv values clipped to half their true magnitude; if any pair of robots violates the minimum safety distance during the transient, the safety proof's main premise fails. Also, start a team whose leader set does not form a rigid subgraph and observe whether the maintenance graph ever becomes rigid—the paper's construction assumes leader rigidity as a starting point.","tokens_in":20644,"feed_emoji":"🤖","tokens_out":3838,"duration_ms":47602,"temperature":0.7,"pith_summary":"The paper targets a gap in multi-robot formation control: most methods assume a rigid communication graph already exists, but in practice the graph starts non-rigid and must be built under limited sensing. The authors propose a leader–follower architecture built on an interleaved layered tree structure (iLDAG), plus a splay scheme that assigns every robot a unique position so it senses a second 'cross-family' link. They prove that adding one such link per follower turns the tree graph into a rigid maintenance graph, satisfying Laman's rigidity conditions. The distributed controller combines hierarchical second-order consensus with collision-cone control barrier functions, guaranteeing collision-free motion and finite-time rigidity acquisition. The claim is that this works for heterogeneous nonlinear robots, with followers using only relative measurements.","feed_headline":"Controller turns non-rigid robot swarms rigid without GPS","feed_subtitle":"Distributed CBF-QP builds collision-free rigid graphs with only relative position sensing.","key_machinery":"The central object is the maintenance graph, a selected subset of sensed edges that carries the rigidity guarantee. The key mechanism is the splay scheme: each parent's children are placed uniformly on a circle whose radius decays geometrically across layers, forcing cross-family links to fall within sensing range. The formal engine is Lemma 1 (edge addition preserves rigidity), Lemma 3 (sensing-range bound for acquiring a second parent), and Theorem 4 (a C3BF-QP controller that combines hierarchical consensus with collision-cone barrier functions).","core_discovery":"The paper's central claim is that rigidity does not need to be assumed—it can be acquired. The authors construct a time-invariant spanning forest (iLDAG) and let each follower move toward a splay orbit so that it enters the sensing range of a designated second parent. By Lemma 1, adding this cross-family edge to each follower preserves Laman's rigidity conditions, so the maintenance graph becomes rigid with the minimal 2n−3 edges. The controller in Theorem 4 guarantees finite-time acquisition of these links while a collision-cone CBF safety filter keeps every pairwise distance above the safety radius, even when agents accelerate and have different sensing ranges.","pith_inferences":["A hidden practical bottleneck is that the leader subgraph is assumed rigid and leaders are assumed to know their absolute positions; if the leaders themselves lose global sensing, the hierarchy's top layer needs a separate rigidity mechanism.","Assumption 4's recursively accumulated bounds Bp and Bv require reliable multi-hop communication of parent state; in lossy networks, underestimated bounds would undermine the safety proof—an empirical stress test would clarify how much slack the CBF formulation has.","The splay geometry essentially pre-plans each follower's target position; a natural extension would be to handle dynamic obstacles or formation reconfiguration after the graph becomes rigid, which the paper does not address.","The 2D rigidity argument relies on Laman's theorem; extending to 3D would require replacing it with generic rigidity in R^3 and adjusting the orbit geometry accordingly."],"forward_implications":["If the central claim holds, multi-robot systems can start from a non-rigid, minimally connected graph and still form a rigid formation, removing a common hidden assumption in formation control.","Rigidity is achieved with the minimum number of edges (2n−3), reducing communication and sensing overhead in large teams.","The safety filter extends collision-cone barrier functions to accelerating obstacles, not just constant-velocity ones, which broadens applicability to real robots.","Controller gains are selected from explicit stability conditions rather than trial and error, simplifying deployment.","Followers never need their own global position or velocity, making the method suitable for GPS-denied indoor or underground environments."],"fun_headline_variants":["Rigidity on the fly: safe graphs without GPS","No initial rigidity needed: safe formation control","Acquire rigid graphs with limited sensing, no GPS","Finite-time rigid graphs with collision safety","From loose to rigid: safe control under limited sensing"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The whole safety and convergence proof leans on Assumption 4: every follower must receive from its parent recursively accumulated bounds on all upstream relative positions and velocities—if those bounds are wrong, delayed, or unavailable, the collision-avoidance and rigidity guarantees collapse.","fun_headline_variants_meta":{"raw":{"variants":["Rigidity on the fly: safe graphs without GPS","No initial rigidity needed: safe formation control","Acquire rigid graphs with limited sensing, no GPS","Finite-time rigid graphs with collision safety","From loose to rigid: safe control under limited sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000163,"raw_usage":{"total_tokens":1039,"prompt_tokens":664,"completion_tokens":375,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":408,"completion_tokens_details":{"reasoning_tokens":316}},"tokens_in":408,"tokens_out":375,"duration_ms":4425,"temperature":1.0,"reasoning_tokens":316,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T07:19:22.742692+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the controller with a single follower whose parent deliberately reports Bp/Bv values clipped to half their true magnitude; if any pair of robots violates the minimum safety distance during the transient, the safety proof's main premise fails. Also, start a team whose leader set does not form a rigid subgraph and observe whether the maintenance graph ever becomes rigid—the paper's construction assumes leader rigidity as a starting point.","supporting_citations":[],"review_version":2}