{"id":"4da0d311-1d17-4e16-b616-e23932cb49a7","arxiv_id":"2607.05544","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Under identical nonlinear PID execution, a finite-horizon control-effort-aware trajectory planner reduces tracking error, corrective torque, and execution cost versus classical kinematic planners on a simplified UR5.","lead":"A control-aware offline trajectory planner that uses manipulator dynamics and actuator effort cuts tracking error and corrective torque versus cubic, quintic, and trapezoidal planners on a simulated UR5 arm. The fair same-controller comparison shows that kinematic smoothness alone does not guarantee efficient nonlinear execution.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"Unreported free parameters (Q,R,midpoint,PID gains) make the reported superiority under fixed nonlinear PID execution non-reproducible and potentially non-robust.","rationale":"The reader correctly isolates the midpoint-linearization-plus-unreported-parameters assumption as the weakest link. My stress-test simply sharpens the same point: without the numerical free parameters the quantitative superiority in Table I is non-reproducible, and a modest re-tuning of Q/R or of the linearization point can in principle erase or reverse the advantage under the fixed PID. That does not invalidate the experimental design (identical nonlinear plant + controller is a genuine strength), nor does it make the qualitative claim false; it does keep the paper at CONDITIONAL until either the parameters are released or a sensitivity study shows the ranking is stable. No stronger internal inconsistency was found; the math of the QP and the evaluation protocol are standard and readable. Hence the verdict remains CONDITIONAL and agreement with the reader is full.","tokens_in":7982,"tokens_out":648,"duration_ms":5987,"concrete_test":"Obtain or re-implement the exact Q,R, midpoint (¯q,¯u), PID gains and Qexec/Rexec used for Table I; re-run the identical nonlinear closed-loop experiment while (i) scaling the planning R by factors {0.1,1,10} and (ii) shifting the linearization point 20 % of the way toward the start and end configurations. If any of the four headline metrics (RMS error, RMS Δτ, integrated |Δτ|, Jexec) reverse ranking versus cubic/quintic for any of those six settings, the reported reductions are tuning-dependent and the claim weakens.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim rests on Table I / §VI metrics (RMS error 2.825e-4 rad, RMS Δτ 1.893 Nm, Jexec 3.622, ~48–62 % cost reductions) obtained when the midpoint-linearized finite-horizon QP trajectory is tracked by the same feedforward-assisted PID under full nonlinear RK4 dynamics. Those numbers are produced by a specific, completely unreported choice of the planning weights Q,R, the midpoint pair (¯x,¯u) used for Ac,Bc (eqs. 6–11), the PID gains Kp,Ki,Kd, and the execution weights Qexec,Rexec. Because the classical planners are purely kinematic and therefore independent of Q,R, any favorable tuning of those free parameters can systematically lower the open-loop residual that the fixed PID must correct, inflating the apparent advantage of the “control-aware” planner. The paper never states the numerical values, never reports a sensitivity sweep, and never quantifies linearization residual along the realized path; consequently the quantitative superiority that underpins the claim “kinematic smoothness alone does not ensure dynamically efficient execution” cannot be verified or stress-tested from the manuscript alone.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper argues that classical kinematic trajectory planners (cubic, quintic, trapezoidal) can produce inefficient nonlinear closed-loop execution, and proposes a control-aware finite-horizon optimal planner that penalizes state deviation and actuator effort after midpoint linearization of the manipulator dynamics. A unified evaluation framework executes all planners under identical nonlinear 3-DoF UR5 dynamics, feedforward-assisted PID structure, gains, and actuator saturation. Simulations report lower RMS tracking error, RMS corrective torque, integrated corrective activity, and closed-loop execution cost for the proposed planner (Table I: roughly 48% cost reduction vs cubic and >62% vs quintic/trapezoidal), supporting the claim that kinematic smoothness alone does not ensure dynamically efficient execution.","tokens_in":8288,"tokens_out":1340,"duration_ms":15974,"significance":"If the quantitative superiority holds under transparent, reproducible settings, the work usefully isolates trajectory-generation effects from controller and model confounds—a comparison design that is often missing in the literature. The main algorithmic idea (finite-horizon quadratic regulation on a midpoint-linearized discrete model, solved offline with quadprog) is standard rather than a new LQR/MPC/iLQR method, as the authors note; the contribution is primarily the controlled nonlinear evaluation and the empirical demonstration. Strengths include a clear shared execution metric suite (RMS error, RMS Δτ, J_Δτ, J_exec) and an internally consistent comparison protocol. The result is of moderate practical interest for industrial PTP planning, but its impact depends on reproducibility of free parameters and on whether the advantage survives beyond a single simplified simulation scenario.","major_comments":[{"comment":"§V and Table I: The central quantitative claims (RMS error 2.825e-4 rad, RMS Δτ 1.893 Nm, J_exec 3.622, and the 28–62% reductions) depend on free parameters that are never stated: planning weights Q,R; execution weights Q_exec,R_exec; PID gains K_p,K_i,K_d; and the midpoint pair (x̄,ū) used for A_c,B_c in eqs. (6)–(11). Classical planners do not depend on Q,R, so favorable tuning of those weights (and of the linearization point) can systematically reduce open-loop residual seen by the fixed PID and inflate the reported advantage. Without numerical values, a sensitivity study, or released code, the Table I superiority is not reproducible and cannot be stress-tested.","section":null},{"comment":"§III.B, eqs. (6)–(11) and §VI: The paper’s load-bearing modeling assumption is that a single midpoint linearization plus ZOH is accurate enough over large PTP motions that the resulting U* remains near-dynamically efficient under full nonlinear RK4 execution. No linearization residual, trajectory-wise model mismatch, or comparison against initial-point / multi-segment linearization is reported. If the linear proxy is poor along the path, the planned reference need not be a good open-loop input, and superiority under fixed PID may shrink or reverse. A residual plot or multi-point linearization ablation is needed to support the midpoint strategy as claimed in the abstract and introduction.","section":null},{"comment":"§V Experimental Setup vs Abstract/§VII: Experiments describe a single fixed-duration PTP task (T_f=2 s, N=100) on a reduced 3-DoF UR5, yet the abstract claims “substantial reductions … across all evaluated scenarios.” Only one motion class, one horizon, and simulation-only results are shown. The claim that kinematic smoothness does not ensure dynamic efficiency is therefore supported only for this narrow setting. At minimum, multiple start/goal pairs, durations, or a second manipulator model (and explicit acknowledgment of the single-scenario scope) are required before the broad wording is justified.","section":null},{"comment":"§V, eqs. (34)–(36): The feedforward term Δτ_ff is not defined. Whether it is inverse-dynamics of the planned (q, q̇, q̈), gravity compensation only, or something else materially affects how much “corrective” effort is attributed to the planner versus the controller. Because the paper’s thesis is reduced corrective demand under identical closed-loop structure, the exact composition of Δτ_cmd must be specified so that differences can be attributed to trajectory generation rather than inconsistent feedforward construction across planners.","section":null}],"minor_comments":[{"comment":"Notation inconsistency: §III.A uses n-DoF and τ = D(q)q̈ + …, while experiments use a “reduced 3-DoF UR5”; state the exact joint subset and inertial parameters used.","section":null},{"comment":"Eq. (5) still writes τ inside f(x,u) after introducing gravity-compensated u in (2)–(3); clarify whether the linearization is about u or τ.","section":null},{"comment":"Fig. 1–7 captions are minimal; axis units and which joint is shown would improve readability.","section":null},{"comment":"Related work cites standard sources but underplays recent torque-/energy-aware trajectory optimization; a short positioning against iLQR/DDP offline generators would clarify novelty of the evaluation framework versus the planner.","section":null},{"comment":"Typographical: “PROBLEMFORMULATION”, “RELATEDWORK”, “CONTROL-AWAREOPTIMAL…” missing spaces in section headings; “F ormulation” / “F . Baseline” spacing artifacts.","section":null},{"comment":"Abstract “across all evaluated scenarios” should be aligned with the single-task experimental description once scope is clarified.","section":null}],"recommendation":"major_revision","confidential_remarks":"Algorithmically this is close to textbook finite-horizon LQR on a linearized manipulator with offline QP; the real contribution is the fair closed-loop comparison design. I would accept after major revision if parameters are disclosed, linearization residual is shown, and multi-scenario results (or carefully narrowed claims) are added. Fit is reasonable for a robotics applications / methods venue; less so for a top theory venue. No integrity red flags beyond non-reproducible free parameters."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing: under a carefully locked-down nonlinear PID execution setup on a simplified 3-DoF UR5, their finite-horizon control-effort-aware planner (midpoint linearization + QP with terminal constraint) consistently beats cubic, quintic, and trapezoidal references on RMS error, corrective torque, and closed-loop cost. The isolation of the planner is the real contribution; the optimizer itself is not new.\n\nWhat they do well is the evaluation design. All four trajectories run on the same nonlinear model, same feedforward-assisted PID, same gains, same saturation, same RK4. That is cleaner than a lot of prior “our optimal method vs classical” papers that change the controller or the plant. The metrics are sensible (RMS e, RMS Δτ, integrated |Δτ|, J_exec), Table I is clear, and the observation that quintic can look worse than cubic under dynamics is worth having on the record. Math is standard LQR-style finite-horizon QP; citations are appropriate and not padded.\n\nSoft spots, in proportion: no hardware, reduced DoF, and—most importantly—the free parameters that actually produce Table I (planning Q,R, midpoint pair, PID gains, execution weights) are never stated. Without them you cannot re-implement the numbers or check sensitivity to linearization residual. Midpoint linearization for large PTP is a reasonable engineering choice, not a deep result; if the linear model is a poor proxy the advantage can shrink. Novelty is incremental relative to textbook LQR/MPC trajectory generation; they correctly disclaim being a new solver.\n\nThis is for people who care about offline industrial trajectory generation under fixed tracking controllers. A serious referee should see it: the experimental logic is sound enough to deserve review, even if the revision will demand parameter tables, sensitivity, and preferably code. I would skim it for the evaluation protocol and the quintic-vs-dynamics point; I would not cite the percentages until they are reproducible.","headline":"Solid same-controller sim comparison showing control-aware offline planning beats cubic/quintic/trap under fixed PID; useful isolation of planner effects, but unreported Q/R/PID/midpoint make the numbers non-reproducible.","tokens_in":8947,"tokens_out":504,"would_cite":false,"duration_ms":4406,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"When robot trajectories are planned with dynamics and actuator effort in mind, closed-loop execution needs less corrective torque and costs less than classical smooth profiles under identical nonlinear control.","keywords":["trajectory planning","optimal control","robotic manipulators","nonlinear execution","control-aware planning","corrective torque","midpoint linearization"],"falsifier":"Repeat the exact same point-to-point task on the same nonlinear UR5 model but replace the midpoint linearization with either successive local linearizations along the trajectory or a fully nonlinear optimizer; if the reported cost and torque reductions disappear or reverse relative to the cubic/quintic baselines under identical PID, the midpoint-approximation claim fails.","tokens_in":8778,"feed_emoji":"🤖","tokens_out":996,"duration_ms":7886,"temperature":0.7,"pith_summary":"Classical joint-space trajectory planners (cubic, quintic, trapezoidal) are purely kinematic: they ignore manipulator dynamics and the torque a feedback controller will later have to apply. This paper shows that such kinematic smoothness can still force a nonlinear robot to expend large corrective effort when the same trajectory is tracked under realistic dynamics. The authors introduce a control-aware planner that linearizes the manipulator at the motion midpoint, formulates a finite-horizon quadratic program that penalizes both state error and actuator effort, and produces an open-loop reference that already anticipates the closed-loop cost. All four planners are then executed on the identical nonlinear 3-DoF UR5 model, with the same feedforward-assisted PID, gains, sampling, and torque limits, so that any performance difference can be attributed solely to how the reference was generated. Simulations report that the control-aware references cut RMS tracking error, RMS corrective torque, integrated corrective activity, and cumulative execution cost by roughly 28–62 % relative to the classical baselines, demonstrating that kinematic smoothness alone does not guarantee dynamically efficient execution.","feed_headline":"Dynamics-aware paths cut robot corrective torque by up to 41%","feed_subtitle":"Same PID, same nonlinear UR5: control-aware references beat cubic, quintic, and trapezoidal profiles on cost and error.","key_machinery":"Midpoint linearization of the nonlinear manipulator dynamics about a single operating pair (¯x, ¯u), followed by ZOH discretization and a finite-horizon quadratic program that minimizes a stacked cost of regulation error and control effort subject to a hard terminal-state constraint and actuator bounds; the resulting open-loop control sequence is then tracked by a fixed nonlinear PID under full rigid-body dynamics.","core_discovery":"Under identical nonlinear dynamics, identical feedforward-assisted PID structure and gains, and identical actuator constraints, a finite-horizon control-aware trajectory that incorporates midpoint-linearized manipulator dynamics and actuator-effort penalties produces lower tracking error, lower corrective torque, and lower closed-loop execution cost than cubic, quintic, or trapezoidal kinematic profiles on a simplified 3-DoF UR5, showing that higher-order kinematic smoothness is not sufficient for dynamically efficient nonlinear execution.","pith_inferences":["If the midpoint linearization is the dominant modeling error, multi-segment or successive linearization schemes should further widen the performance gap over pure kinematic profiles.","The same unified evaluation protocol could be used to rank other popular generators (B-splines, time-optimal bang-coast-bang, learning-based planners) without confounding controller differences.","Because the cost reductions appear even under a simple PID, the advantage is likely to persist or grow under more sophisticated trackers that still treat the planned trajectory as an open-loop reference."],"forward_implications":["Industrial robot programmers who currently rely on cubic or quintic joint interpolators can reduce actuator wear and energy simply by swapping the open-loop reference generator while leaving the existing PID tracker unchanged.","Trajectory quality metrics should include closed-loop corrective-torque and execution-cost integrals, not only kinematic smoothness or peak jerk.","Offline finite-horizon quadratic programs that already embed actuator effort can serve as practical drop-in replacements for classical profiles whenever the task is structured point-to-point motion under fixed feedback control.","Comparative studies of trajectory generators become more conclusive when all methods are forced to execute under identical nonlinear dynamics, controller structure, and saturation limits."],"fun_headline_variants":["Control-aware plans cut UR5 torque and error vs cubic quintic","Same PID: dynamics-aware paths beat classical kinematic profiles","Midpoint-linearized trajectories lower closed-loop robot costs","Kinematic smoothness alone fails for efficient nonlinear execution","Finite-horizon control-aware refs reduce tracking error and effort"],"cache_read_input_tokens":128,"weakest_assumption_plain":"A single midpoint linearization of the robot’s nonlinear dynamics remains accurate enough over a large point-to-point motion that the quadratic-program trajectory stays near-optimal when the full nonlinear plant is later closed under PID.","fun_headline_variants_meta":{"raw":{"variants":["Control-aware plans cut UR5 torque and error vs cubic quintic","Same PID: dynamics-aware paths beat classical kinematic profiles","Midpoint-linearized trajectories lower closed-loop robot costs","Kinematic smoothness alone fails for efficient nonlinear execution","Finite-horizon control-aware refs reduce tracking error and effort"]},"model":"grok-4.5","effort":"low","cost_usd":0.005892,"raw_usage":{"total_tokens":1540,"prompt_tokens":789,"num_sources_used":0,"completion_tokens":71,"cost_in_usd_ticks":58920000,"prompt_tokens_details":{"text_tokens":789,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":680,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":789,"tokens_out":71,"duration_ms":5580,"temperature":1.0,"reasoning_tokens":680,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-11T06:02:31.638264+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Repeat the exact same point-to-point task on the same nonlinear UR5 model but replace the midpoint linearization with either successive local linearizations along the trajectory or a fully nonlinear optimizer; if the reported cost and torque reductions disappear or reverse relative to the cubic/quintic baselines under identical PID, the midpoint-approximation claim fails.","supporting_citations":[],"review_version":1}