{"id":"61a2e308-559c-43ae-ad6d-ecf1269dfd76","arxiv_id":"2605.26625","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Develops a probabilistically complete sampling-based planner that learns lower-dimensional Wasserstein ambiguity tubes from data and uses a bandit-based checker to find safe paths under unknown disturbances for linear robotic systems.","lead":"The paper introduces a sampling-based motion planning algorithm that learns Wasserstein ambiguity tubes from trajectory data to ensure chance-constrained safety for robots under unknown random disturbances. A smart generalist might care because it targets a common real-world problem where exact disturbance statistics are unavailable, potentially enabling more reliable autonomous systems.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Wasserstein ambiguity sets from finite trajectory data plus lower-dim projections may fail to contain true joint distribution with stated confidence","rationale":"The reader's weakest_assumption directly identifies the data-to-set and projection steps that must hold for the 'provably safe' label; without the full proofs or the specific concentration constants used, the claim cannot be verified. No stronger internal inconsistency appears from the abstract.","tokens_in":1697,"tokens_out":286,"duration_ms":26216,"concrete_test":"Extract the exact statement and proof of Theorem X (or equivalent) that bounds the radius of the projected ambiguity sets; recompute the radius on a held-out trajectory dataset of size equal to the paper's training set and check whether the empirical coverage probability falls below the claimed 1-δ.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The central claim requires that finite trajectory samples yield Wasserstein balls whose union (the ambiguity tube) contains the true state-distribution trajectory with high probability, and that independent lower-dimensional tubes can be substituted while preserving the joint chance-constraint guarantees used by the sampling-based planner. Both steps rest on standard Wasserstein concentration results that assume i.i.d. samples and that marginal projections do not inflate the radius needed for joint coverage; neither assumption is automatically true for closed-loop robotic trajectories or when obstacles induce state-dependent disturbance statistics.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims to present a provably safe sampling-based motion planning algorithm for linear or linearizable robotic systems under unknown disturbances. Safety is encoded as chance constraints. The method learns a Wasserstein ambiguity tube (sequence of ambiguity sets) from system trajectory data that contains the true state-distribution trajectory with high confidence; this tube is used to grow a probabilistically complete sampling-based planning tree. The authors further claim that replacing a single high-dimensional tube with several lower-dimensional ones reduces conservatism and improves scalability, and that a bandit-based validity checker boosts empirical performance without losing probabilistic completeness. Case studies are said to show outperformance over state-of-the-art methods in cluttered environments.","tokens_in":1865,"tokens_out":589,"duration_ms":39698,"significance":"If the central guarantees on the ambiguity tube and its lower-dimensional projections hold, the work would supply a data-driven route to formal safety certificates for sampling-based planners under distributionally uncertain disturbances. The combination of Wasserstein ambiguity sets with probabilistic completeness and the proposed dimensionality-reduction and bandit techniques would be a concrete advance for safe robotics in unstructured settings.","major_comments":[{"comment":"The central claim that finite trajectory samples yield Wasserstein balls whose union contains the true joint state-distribution trajectory with high probability rests on concentration results that require i.i.d. samples; closed-loop robotic trajectories collected under feedback control and obstacle-induced state-dependent disturbances are unlikely to satisfy this assumption, yet no correction or robustness argument is supplied for the data-generating process.","section":"Abstract"},{"comment":"The substitution of independent lower-dimensional ambiguity tubes for a joint high-dimensional tube is asserted to preserve the required probabilistic guarantees for the chance constraints used by the planner; however, marginal projections generally inflate the radius needed for joint coverage, and no explicit radius-adjustment rule or proof that the joint chance-constraint satisfaction is retained appears in the provided description.","section":"Abstract"},{"comment":"The bandit-based validity checker is claimed to increase empirical performance without sacrificing probabilistic completeness, but the interaction between the checker and the ambiguity-tube safety certificates is not shown to maintain the overall high-confidence containment property when invalid samples are rejected.","section":"Abstract"}],"minor_comments":[{"comment":"The abstract states that the algorithm 'outperforms state-of-the-art methods' but does not name the specific baselines or report quantitative metrics (success rate, computation time, conservatism) in the summary paragraph.","section":null},{"comment":"Notation for the Wasserstein ambiguity tube (sequence of sets, radius, confidence level) is introduced without an accompanying equation or definition in the abstract, making it difficult to trace how the tube is constructed from data.","section":null}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback. We respond to each major comment below and indicate planned revisions to address the concerns.","responses":[{"response":"The concentration results invoked in the paper are stated under the i.i.d. sampling assumption for the collected trajectories. The manuscript describes data collection via multiple independent rollouts. We agree that dependence induced by feedback and obstacles merits explicit treatment. In revision we will add a dedicated paragraph in the data-collection and theoretical sections discussing this assumption, citing mixing-time results that extend Wasserstein concentration to weakly dependent processes, and noting the practical regime in which the i.i.d. approximation remains valid.","revision_made":"yes","referee_comment":"[Abstract] The central claim that finite trajectory samples yield Wasserstein balls whose union contains the true joint state-distribution trajectory with high probability rests on concentration results that require i.i.d. samples; closed-loop robotic trajectories collected under feedback control and obstacle-induced state-dependent disturbances are unlikely to satisfy this assumption, yet no correction or robustness argument is supplied for the data-generating process."},{"response":"The dimensionality-reduction argument in the manuscript relies on a per-marginal coverage guarantee combined with a union-bound adjustment of the failure probability. We will insert an explicit radius-correction formula (derived from the union bound) together with a short proof that the resulting joint chance constraint remains satisfied at the target level. This material will be placed in the section on lower-dimensional tubes.","revision_made":"yes","referee_comment":"[Abstract] The substitution of independent lower-dimensional ambiguity tubes for a joint high-dimensional tube is asserted to preserve the required probabilistic guarantees for the chance constraints used by the planner; however, marginal projections generally inflate the radius needed for joint coverage, and no explicit radius-adjustment rule or proof that the joint chance-constraint satisfaction is retained appears in the provided description."},{"response":"The checker only discards candidate nodes that already violate the tube-derived certificates; it does not alter the underlying high-probability containment of the true distribution inside the tube. We will augment the probabilistic-completeness theorem with an additional lemma that formalizes this separation, showing that rejection decisions preserve the original containment probability. The expanded argument will appear in the analysis of the validity checker.","revision_made":"yes","referee_comment":"[Abstract] The bandit-based validity checker is claimed to increase empirical performance without sacrificing probabilistic completeness, but the interaction between the checker and the ambiguity-tube safety certificates is not shown to maintain the overall high-confidence containment property when invalid samples are rejected."}],"tokens_in":1401,"tokens_out":553,"duration_ms":41920,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core contribution is a sampling-based planner that learns a sequence of Wasserstein ambiguity sets from system trajectories to enforce chance constraints without knowing the disturbance distribution. It adds the use of multiple lower-dimensional tubes to cut conservatism and a bandit-based validity checker to speed up tree growth while keeping probabilistic completeness. The case studies reportedly show better performance than prior methods in cluttered workspaces with tight safety levels.\n\nThe approach is a reasonable extension of distributionally robust ideas into motion planning. The lower-dimensional tube idea and the bandit checker are concrete engineering steps that could matter for scalability.\n\nThe main weakness is the central assumption that finite trajectory samples produce ambiguity tubes containing the true joint state-distribution trajectory with the claimed high probability, and that independent lower-dimensional tubes can be swapped in without breaking the joint guarantees. Robotic trajectories are closed-loop and often state-dependent due to obstacles, so they violate the i.i.d. conditions behind standard Wasserstein concentration bounds. The abstract gives no indication that the paper tests or adjusts for this. If the full proofs do not address the projection step rigorously, the safety claims weaken.\n\nThis is aimed at researchers in robust robot motion planning who already work with sampling-based methods and distributionally robust optimization. A reader already familiar with Wasserstein DRO would get the most out of the specific integration and the empirical scaling claims.\n\nI would send it to review to check whether the proofs close the gap on the data and projection issues, but the current evidence from the abstract alone does not make the guarantees look solid.","headline":"The paper combines Wasserstein ambiguity tubes with sampling-based planning and lower-dimensional projections for chance-constrained safety under unknown disturbances, but the guarantees rest on data assumptions that may not hold for closed-loop robotic trajectories.","tokens_in":2317,"tokens_out":391,"would_cite":false,"duration_ms":21993,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Trajectory data yields Wasserstein ambiguity tubes that let sampling-based planners meet chance constraints under unknown disturbances.","keywords":["motion planning","sampling-based planning","chance constraints","Wasserstein ambiguity sets","unknown disturbances","probabilistic safety","data-driven planning","robotic systems"],"falsifier":"A concrete counter-example in which the true disturbance distribution lies outside the learned ambiguity tube yet the planner still declares a trajectory safe, or an instance where the lower-dimensional projection loses the original chance-constraint guarantee.","tokens_in":2613,"feed_emoji":"🤖","tokens_out":709,"duration_ms":34643,"temperature":0.7,"pith_summary":"The paper develops a sampling-based motion planner that remains safe for robots whose dynamics are subject to random disturbances whose distribution is unknown. It learns a Wasserstein ambiguity tube directly from observed trajectories; the tube is a sequence of ambiguity sets guaranteed to contain the true state-distribution trajectory with high confidence. This tube is then used to grow a planning tree that respects state, control, and obstacle constraints while satisfying the required chance constraints. Replacing one high-dimensional tube with several lower-dimensional ones reduces conservatism, and a bandit-based validity checker speeds up tree expansion without losing probabilistic completeness. The result is an algorithm that finds valid plans in cluttered workspaces under strict safety thresholds where prior methods fail.","feed_headline":"Data tubes let sampling planners meet chance constraints under unknown noise","feed_subtitle":"Wasserstein ambiguity tubes learned from trajectories grow probabilistically complete trees that respect safety thresholds in cluttered spac","key_machinery":"Wasserstein ambiguity tube: a sequence of data-driven ambiguity sets that bounds the state-distribution trajectory with high confidence and is projected to lower dimensions to enforce chance constraints inside the sampling-based planner.","core_discovery":"Trajectory data from the system is used to construct a Wasserstein ambiguity tube, defined as a sequence of ambiguity sets that contains the trajectory of the system's state distribution with high confidence. This tube is incorporated into a probabilistically complete sampling-based planner that grows a tree respecting state and control constraints as well as arbitrary-shaped obstacles. Learning multiple lower-dimensional ambiguity tubes in place of a single high-dimensional tube reduces conservatism and improves scalability, while a bandit-based validity checker further increases empirical speed without affecting the completeness guarantee.","pith_inferences":["The approach suggests that similar ambiguity-tube constructions could be tested on physical hardware by collecting real trajectory data under controlled disturbances.","If the tube construction is combined with online data collection, the planner could adapt its safety margins during execution.","The lower-dimensional projection technique may generalize to other distributionally robust planning problems beyond sampling-based methods."],"forward_implications":["Robots can generate motion plans that satisfy chance constraints even when the disturbance distribution is completely unknown.","High-dimensional systems become tractable because multiple lower-dimensional tubes replace a single conservative high-dimensional tube.","Probabilistic completeness of the tree-growing algorithm is retained while empirical runtime improves via the bandit checker.","The same data-driven tube construction applies to both linear and linearizable dynamics subject to state, control, and workspace constraints."],"fun_headline_variants":["Wasserstein tubes support safe sampling plans under disturbances","Multiple lower dim tubes reduce conservatism in planning","Trajectory data builds tubes for chance constrained safety","Bandit checker aids validity in sampling based planners","Provably safe motion planning via learned ambiguity tubes"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The collected trajectory data is representative enough that the resulting Wasserstein ambiguity sets contain the true state-distribution trajectory with the claimed high confidence, and that lower-dimensional projections preserve the necessary probabilistic guarantees.","fun_headline_variants_meta":{"raw":{"variants":["Wasserstein tubes support safe sampling plans under disturbances","Multiple lower dim tubes reduce conservatism in planning","Trajectory data builds tubes for chance constrained safety","Bandit checker aids validity in sampling based planners","Provably safe motion planning via learned ambiguity tubes"]},"model":"grok-4.3","cost_usd":0.00571,"raw_usage":{"total_tokens":2709,"prompt_tokens":635,"num_sources_used":0,"completion_tokens":60,"cost_in_usd_ticks":57099500,"prompt_tokens_details":{"text_tokens":635,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2014,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":635,"tokens_out":60,"duration_ms":9244,"temperature":1.0,"reasoning_tokens":2014,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-29T17:27:55.005065+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A concrete counter-example in which the true disturbance distribution lies outside the learned ambiguity tube yet the planner still declares a trajectory safe, or an instance where the lower-dimensional projection loses the original chance-constraint guarantee.","supporting_citations":[],"review_version":1}