{"id":"596c2164-471b-4462-b0f6-637e3fed4b18","arxiv_id":"2501.15272","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A multi-drone cable-suspended transport system that plans safe, agile trajectories in real time and tracks them without payload or cable sensors is validated in simulation and flight tests.","lead":"This paper presents a planning and control system that lets a team of drones carry a cable-suspended payload through cluttered spaces with real-time replanning. It works without measuring the payload or cable states, using only each drone's own sensors, and is demonstrated in experiments with three drones.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The no-feedback control claim is only validated for aerial-robot tracking; the actual payload trajectory and cable tautness are never measured, so 'agile transportation of the payload' and the robustness results in Tables IV-VI are not yet supported.","rationale":"The reader's weakest assumption targets the planner's use of soft, finitely sampled constraints (Eq. 57-59). That is a genuine correctness risk, but the experiments provide some empirical bounds on it: the tested trajectories were executed without reported collisions. A more central gap is the payload itself. The paper's signature contribution is removing payload and cable sensing and payload closed-loop control, yet no reported metric measures the payload. The planned payload trajectory p(t) is part of the flat output, but execution errors in the aerial robots can perturb the payload's zero dynamics and cable tautness, and no theorem bounds this perturbation. This is an absence of evidence rather than a demonstrated contradiction, so it does not warrant rejection; it strengthens the case for a conditional verdict. The reader's concern and this one partially overlap in that both say the full claim is not yet verified, but they identify different missing pieces: the reader emphasizes planner certification, while this concern emphasizes controller validation. Adding one payload-state measurement experiment, or a stability/slack analysis, would settle the issue.","tokens_in":26094,"tokens_out":6632,"duration_ms":71589,"concrete_test":"Repeat Scenario 2 (agile circle, Table IV) with an external motion-capture marker or high-speed vision mounted on the payload, while keeping the controller feedback-free, and log cable angles or estimated tension along each run. Report payload-position RMSE and maximum error versus the planned p(t), plus the minimum estimated cable tension. If payload RMSE stays comparable to the reported robot RMSE (e.g., within about 10 cm) and no cable approaches slack (F_n < F_min), the no-feedback claim is supported; if the payload drifts or cables slacken, the controller needs payload-state feedback or a stability certificate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section VI removes payload and cable states from the control loop (Eq. 63-66) and estimates cable force via INDI. The planner's flat output Z includes payload position p and cable parameters ξ_n, and the controller tracks p_n = p + lρ_n for each robot (Eq. 4), but there is no feedback or stability analysis for the payload subsystem. With three taut cables, small robot tracking errors can make the exact intersection condition |p - p_n| = l inconsistent, so the taut-mode model (Sec. III-B) may be violated by cable slack; slack is explicitly identified in Sec. III-B as dangerous to control. The experimental RMSEs reported in Tables IV, V, VI and Figure 9 are robot-trajectory errors (Figure 9C shows the second robot's states), not payload position or cable-direction errors. Thus the experiments do not establish that the payload follows the planned agile trajectory, that cables remain taut with F_n ≥ F_min, or that robustness to non-point-mass payloads and mass uncertainty holds for the transported object itself. The headline claim therefore risks being a statement about robot formation tracking rather than payload transportation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a complete planning-and-control framework for multiple aerial robots transporting a cable-suspended payload. The planner uses a differentially flat formulation based on an extended flat output (payload position, cable angles, cable force, yaw), a CMINCO-based sparse spatio-temporal trajectory representation, and an unconstrained optimization with penalty terms for obstacle avoidance, reciprocal avoidance, dynamic feasibility, and payload-coupling dynamics. The controller is a fully distributed two-loop INDI scheme that does not measure payload or cable states and instead estimates cable forces online. The claims are validated in benchmark comparisons against a state-of-the-art kinodynamic planner, in ablation studies, and in real experiments with three 320 g aerial robots carrying payloads up to 200 g at accelerations up to 9.18 m/s^2. The paper positions itself as the first scheme to achieve real-time, safe, agile transportation of a cable-suspended payload in complex environments without any payload or cable state feedback.","tokens_in":26375,"tokens_out":7340,"duration_ms":69339,"significance":"If the claims are fully supported, this would be a substantial contribution to multi-robot aerial transportation: the system is end-to-end, runs in real time, and the hardware experiments demonstrate aggressive maneuvers with a practical three-robot platform. The flatness derivation, the use of INDI for cable-force compensation, and the open-source code are concrete strengths. The reported replanning times (50-162 ms) and the narrow-gap and S-turn experiments are compelling evidence of practical utility. However, the central claims of 'collision-free and dynamically feasible trajectories' and of 'agile transportation of the payload' are stronger than what the current optimization formulation and the reported experimental metrics actually establish. The paper is therefore valuable but needs additional verification or claim softening before the headline results can be accepted at face value.","major_comments":[{"comment":"The claim in the Abstract and §I that the planner produces a 'collision-free and dynamically feasible' trajectory is not supported by the optimization formulation as written. The hard constraints in Eq. (52f) are replaced by a finite trapezoidal-rule penalty (Eqs. 58-59), and the cable-clearance cost uses only K=7 sampled points per cable (Table II). Consequently, constraint violation between quadrature nodes or between cable samples is neither prevented nor detected. In §VII-A, 'success' is defined as generating a dynamically feasible trajectory, but no feasibility check is described. Please report, for the benchmark and experimental trajectories, the maximum violation of Eqs. (20)-(22), (34), (37), (40), and (43) on a fine evaluation grid, or use a formulation with a certified feasibility guarantee.","section":"§V-C3, Eq. (57)-(59), Table II"},{"comment":"The equality constraint that makes the extended flat output Z dynamically consistent with the payload is enforced only through the soft penalty Jd with a finite weight λd. The flatness maps in §IV-A are derived under the exact equation Eq. (47); if the optimized Z has residual Jd > 0, the robot states and inputs computed by these maps are not the actual states and inputs of a feasible MARTS trajectory. The paper does not report the residual value of Jd for any benchmark or experiment. Please report Jd after optimization and either impose Eq. (47) as a hard constraint or provide a bound on the feasibility error induced by the finite penalty weight.","section":"§IV-E, Eq. (47)-(48)"},{"comment":"All reported tracking errors are for the aerial robots (Fig. 9C shows robot 2's states), not for the payload position, cable direction, or cable tension. The paper's central experimental claims—agile transportation of the payload, maintenance of the taut mode (F_n ≥ F_min, Eq. 51), and robustness to non-point-mass payloads—are therefore not directly validated. Since the controller has no payload feedback, the observed robot tracking could in principle coexist with substantial payload deviation or cable slack, and the paper itself notes in §III-B that slack transitions are dangerous to control. Please add payload motion-capture or a high-fidelity estimate of the payload trajectory from the measured robot states, and report payload trajectory error and estimated cable forces along the trajectory, or explicitly weaken the claims to robot-formation tracking.","section":"§VI, Tables IV-VI, Fig. 9"}],"minor_comments":[{"comment":"The symbol m in the cable-force term of Eq. (8) and Eq. (16b) should be m_n (the robot mass) to match Eq. (2b) and Table I; as written, m is undefined in that context.","section":"Eq. (8), Eq. (16b)"},{"comment":"The penalty weights λτ and λς in Table II are not defined in Eq. (61), and the thrust-penalty weight λf appearing in Eq. (61a) has no matching entry in Table II; the notation λT/λZ in Eq. (52a) also differs from the lowercase forms in the table.","section":"Table II, Eq. (61)"},{"comment":"The statement that 'the bounds of the constraints considered in both methods are set to be the same as listed in Tab II' is imprecise, since Table II also lists many non-constraint parameters such as control gains and filter settings.","section":"§VII-A"},{"comment":"The caption contains the typo '2th aerial robot'; it should read '2nd aerial robot'.","section":"Fig. 9 caption"},{"comment":"Please clarify how the hover condition preceding the mass estimation is determined and how W=1000 (Table II) is used in the averaging, since Eq. (74) does not explicitly define the window over which the sequence is collected.","section":"§VI-C, Eq. (74)"}],"recommendation":"major_revision","confidential_remarks":"The engineering contribution is real and the hardware experiments are impressive, but the headline claims of certified collision-free/dynamically feasible trajectories and of validated payload transportation exceed what the current soft-constraint formulation and robot-only error metrics support. The gaps are closable with additional verification (dense-grid constraint violation reports, payload-state measurements) or by softening the claims. I do not see grounds for rejection, but the revision must address the feasibility-certification and payload-validation issues."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. This is a genuinely integrated planning and control pipeline, and the real experiments are the core asset: three 320g robots carried up to 200g through S-shaped corridors and a 1m gap, with replanning at 72-162ms and accelerations up to 9.18m/s^2. That is real engineering. The benchmark against Wahba's planner shows a two-order-of-magnitude speedup at equal or better success rate, and the theta-phi-F flatness parametrization, sector-based vectorial cable constraints, CMINCO-based trajectory deformation, and INDI force estimation fit together coherently. The derivations are dense and mostly plausible, and the ablation on the vectorial constraints supports their purpose.\n\nThe soft spots are real but not fatal. First, \"safe\" and \"collision-free\" are not certified: Eq. 57-59 convert infinite continuous-time constraints into trapezoidal-rule penalties with chosen weights and K=7 cable samples. That can be acceptable in practice, but the abstract's guarantee language outruns the math. Second, and more important, the stress-test note is correct: the payload trajectory and cable tautness are never measured. Figure 9 and Tables IV-VI report robot-tracking RMSEs, not payload position or cable-direction errors. The robustness results in Table VI therefore show that the robots still track their planned formation under payload mass errors and non-point payloads; they do not show that the payload itself followed the planned agile trajectory or that cables stayed taut. The model assumes taut cables and explicitly identifies slack as dangerous, so this unmeasured link matters. It does not invalidate the system demo, but it means the headline claims should be softened. Third, minor: no variance/significance statistics, and the promised open-source link is absent. The CMINCO self-citation is fine; it is a published, independently usable prior method.\n\nWho this is for: people building multi-UAV cable-payload systems, and aerial robotics researchers generally. It deserves a serious referee. A major revision should ask for either payload-state measurements (a motion-capture marker on the payload is cheap) or an explicit reframing of the claims to robot formation tracking with payload robustness as an indirect inference.","headline":"A genuine integrated planner+controller for multi-drone cable transport with impressive real flights, but the headline claims outrun the evidence because payload trajectory and cable tautness are never measured and safety is soft-penalty, not certified.","tokens_in":26895,"tokens_out":3016,"would_cite":true,"duration_ms":30251,"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":"A three-robot team can transport a cable-suspended payload through cluttered environments with agile, near-thrust-limit maneuvers, planning each trajectory in real time and tracking it without any measurement or closed-loop control of the…","keywords":["multiple aerial robots","cable-suspended payload","differential flatness","real-time trajectory planning","distributed control","incremental nonlinear dynamic inversion","collision avoidance","unconstrained optimization"],"falsifier":"Take one optimized trajectory from the paper's planner and evaluate every imposed constraint—obstacle distance (via the ESDF), inter-robot distance, thrust magnitude, tilt angle, and body rate—at dense random time instants between the quadrature nodes and with more than $K=7$ cable samples; any violation means the penalty-based transcription does not enforce the claimed safety and feasibility.","tokens_in":25875,"feed_emoji":"🚁","tokens_out":8254,"duration_ms":66655,"temperature":0.7,"pith_summary":"The paper claims that a team of cable-suspended aerial robots can be made to transport a payload safely and aggressively through cluttered environments, and that this requires neither state measurements nor closed-loop control of the payload or cables. It provides a complete scheme: a real-time trajectory planner that respects the kinematic cable constraint and the dynamic coupling between robots and payload, and a fully distributed controller that tracks the resulting agile trajectory using only each robot's own state. If the claim holds, the practical barriers that have kept multi-robot cable-suspended transport at low agility in simple environments are removed: slow planning, centralized force allocation, and dependence on fragile payload feedback. The paper backs the claim with three-robot experiments carrying payloads up to 200 g at accelerations up to 9.18 m/s², replanning in tens of milliseconds.","feed_headline":"Three drones carry a swinging payload through tight spaces at 9 m/s²","feed_subtitle":"Removes payload sensors and closed-loop payload control, enabling agile multi-robot transport in cluttered scenes.","key_machinery":"The load-bearing object is the extended flat-output variable $Z = (p^T, \\xi_1^T, \\psi_1, \\dots, \\xi_N^T, \\psi_N)^T$, with $\\xi_n = (\\theta_n, \\phi_n, F_n)$, where $p$ is the payload position, $\\theta_n$ and $\\phi_n$ are the pitch and azimuth angles of the $n$-th cable, and $F_n$ is the cable tension. Flatness maps derived from this variable convert the flat output into each robot's mass-normalized thrust vector, tilt angle, and body rate, so that trajectory optimization can be performed entirely on the flat-output space. The surrounding machinery is the representation of each robot's position as $p_n = p + l \\rho_n$ (which eliminates the taut-cable kinematic constraint), the piecewise-polynomial trajectory class with sparse spatio-temporal parameters, the diffeomorphisms that remove cable vectorial and temporal constraints, and the trapezoidal-rule transcription that turns infinite continuous-time constraints into finite penalty terms.","core_discovery":"The central discovery is that the multiple-aerial-robot transportation problem has a structure that permits real-time safe and agile planning: represent each robot's position indirectly as the payload position plus the cable direction times the cable length, so the kinematic constraint is automatically satisfied; then use flatness maps to read off each robot's thrust, tilt, and body rate from an extended flat-output variable made of the payload position, per-cable pitch and yaw angles, tension, and yaw angle. With these maps, the paper converts trajectory planning into an unconstrained optimization over sparse spatio-temporal parameters, eliminating cable direction and tension constraints by diffeomorphisms and approximating infinite-time safety and feasibility constraints by trapezoidal-rule penalties. For tracking, it replaces payload and cable feedback with an incremental nonlinear dynamic inversion controller that estimates the actual cable force from motor speeds and estimates payload mass during hover, making the system robust to mass error and non-point-mass payloads. The experimental evidence—up to 9.18 m/s² accelerations, narrow-gap traversal, and 72 ms emergency replanning—supports the claim that this is the first complete scheme to achieve such agile transportation in complex environments without payload sensing.","pith_inferences":["The soft-penalty formulation suggests that for agile multi-robot manipulation, planning with approximate constraints plus a disturbance-rejecting controller may be a more practical route than hard-constrained optimization, at the price of provable safety guarantees.","The same flatness-based pipeline could extend to other underactuated cooperative systems with taut connections (rigid rods, winches) by changing the connection model while keeping the flat-output and penalty-transcription structure.","Because the obstacle constraints are sampled on the cable at $K=7$ points, a thin obstacle passing between two samples could go undetected; a testable extension is to adapt $K$ locally based on obstacle proximity or cable length.","The mass estimation performed at hover could be run continuously during flight, enabling adaptation to payloads that gain or lose mass mid-mission."],"forward_implications":["Emergency retargeting becomes practical: the reported replanning time is about 72 ms, so the system can switch targets mid-flight while keeping a safe distance from obstacles.","The formation can actively deform—contracting or reorienting the cables—to pass through gaps narrower than the formation's nominal width, as demonstrated with a 1 m gap.","Payload and cable feedback can be removed entirely; cable force is reconstructed from motor speeds, and payload mass is estimated during a hover phase, so no sensors need to be attached to the payload or cables.","Tracking error remains bounded under payload mass errors up to ±30% and for non-point-mass payloads such as a carton or water bottle, with RMSE increases of at most about 25% in the reported tests."],"supporting_citations":[{"why":"Establishes differential flatness of the multi-robot cable-suspended system with taut cables, the basis for planning on flat outputs.","marker":"[22]"},{"why":"Provides the state-of-the-art kinodynamic planner used as the benchmark in success-rate and responsiveness comparisons.","marker":"[26]"},{"why":"Supplies the piecewise-polynomial trajectory class whose sparse parameters enable linear-complexity trajectory generation and gradient propagation.","marker":"[41]"},{"why":"Provides the incremental nonlinear dynamic inversion approach used in the outer and inner control loops to compensate forces and moments.","marker":"[39]"},{"why":"Supplies the rotation factorization used to reconstruct desired attitude from thrust direction and yaw.","marker":"[40]"},{"why":"Supplies the constraint transcription that converts infinite continuous-time constraints into integral-type penalty functions.","marker":"[42]"}],"fun_headline_variants":["Agile multi-drone payload transport without payload sensors","Real-time safe trajectory for cable-suspended payloads with drones","No payload sensors needed for agile drone team transport","Drone team carries swinging payload safely and agilely","Multi-robot cable transport: safe and agile without payload feedback"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The planner's safety and dynamic-feasibility guarantees rest on the assumption that checking constraints at discrete sample times, with cable direction and tension ranges enforced as soft penalties of chosen weights, is enough to keep the trajectory safe and executable at all times in between.","fun_headline_variants_meta":{"raw":{"variants":["Agile multi-drone payload transport without payload sensors","Real-time safe trajectory for cable-suspended payloads with drones","No payload sensors needed for agile drone team transport","Drone team carries swinging payload safely and agilely","Multi-robot cable transport: safe and agile without payload feedback"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000431,"raw_usage":{"total_tokens":2261,"prompt_tokens":1070,"completion_tokens":1191,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":686,"completion_tokens_details":{"reasoning_tokens":1112}},"tokens_in":686,"tokens_out":1191,"duration_ms":8329,"temperature":1.0,"reasoning_tokens":1112,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:27:23.202440+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take one optimized trajectory from the paper's planner and evaluate every imposed constraint—obstacle distance (via the ESDF), inter-robot distance, thrust magnitude, tilt angle, and body rate—at dense random time instants between the quadrature nodes and with more than $K=7$ cable samples; any violation means the penalty-based transcription does not enforce the claimed safety and feasibility.","supporting_citations":[{"cited_title":"Dynamics, control and planning for co- operative manipulation of payloads suspended by cables from multiple quadrotor robots,","cited_arxiv_id":null,"evidence_quote":"Establishes differential flatness of the multi-robot cable-suspended system with taut cables, the basis for planning on flat outputs."},{"cited_title":"Kinodynamic Motion Planning for a Team of Multirotors Transporting a Cable-Suspended Payload in Cluttered Environments","cited_arxiv_id":"2310.03394","evidence_quote":"Provides the state-of-the-art kinodynamic planner used as the benchmark in success-rate and responsiveness comparisons."},{"cited_title":"Accurate tracking of aggressive quadrotor tra- jectories using incremental nonlinear dynamic inversion and differential flatness,","cited_arxiv_id":null,"evidence_quote":"Provides the incremental nonlinear dynamic inversion approach used in the outer and inner control loops to compensate forces and moments."},{"cited_title":"Control of quadrotors using the hopf fibration on so (3),","cited_arxiv_id":null,"evidence_quote":"Supplies the rotation factorization used to reconstruct desired attitude from thrust direction and yaw."},{"cited_title":"A computational algorithm for functional inequality constrained optimization problems,","cited_arxiv_id":null,"evidence_quote":"Supplies the constraint transcription that converts infinite continuous-time constraints into integral-type penalty functions."}],"review_version":1}