{"id":"b8454d46-d0a7-416c-a2b3-aa3ee772052d","arxiv_id":"2604.07599","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"SANDO achieves high success rates with collision-free guarantees in dynamic unknown 3D environments via spatiotemporal safe flight corridors, variable-elimination MIQP, and formal analysis under bounded velocity and estimation error.","lead":"SANDO is a trajectory planner for drones in 3D spaces where obstacles move unpredictably and their positions are unknown in advance. It uses risk heat maps for rough paths, time-specific safe corridors that grow only as far as obstacles could reach at each moment, and fast mixed-integer optimization to replan quickly while proving safety under speed and sensing limits.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Safety guarantees require accurate per-timestep worst-case reachable sets under explicit velocity bounds; any mismatch between assumed and actual obstacle dynamics voids the hard constraints.","rationale":"The reader’s weakest assumption directly names the same modeling prerequisites (bounded velocities, accurate reachable sets, bounded estimation error, safe parameter choices). The abstract and strongest claim make the guarantees conditional on these; the concrete test above would falsify or confirm whether the condition is practically fragile.","tokens_in":1829,"tokens_out":381,"duration_ms":44165,"concrete_test":"Re-run the dynamic-environment benchmark suite while injecting a single obstacle that exceeds the planner’s declared velocity bound by 15 % for 2–3 seconds within the horizon; measure both collision rate and success rate for SANDO versus the original bounded case. If collisions appear or feasibility drops sharply, the reachable-set assumption is load-bearing.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The STSFC construction inflates each obstacle only by its reachable set at the specific time layer rather than the full-horizon worst case. This yields tighter corridors and higher feasibility, but the formal collision-free proof (and the MIQP hard constraints) holds only if (1) the velocity bound is known and never exceeded, (2) the estimation error is bounded as stated, and (3) the reachable-set over-approximation is tight enough that the inflated polytope truly contains every possible obstacle position. In unknown environments these quantities are not directly observable; the paper’s simulations and hardware trials use obstacles whose motion obeys the same bounds used by the planner. Consequently the reported “no constraint violations” and “collision-free guarantees” are conditional on the modeling assumptions being exactly satisfied. If an obstacle can accelerate, change direction, or exceed the declared speed even briefly, the corridor may become invalid without the planner detecting it.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript presents SANDO, a trajectory planner for 3D dynamic unknown environments. It combines a heat-map A* global planner that avoids high-risk regions via soft costs, a spatiotemporal safe flight corridor (STSFC) generator that produces time-layered polytopes by inflating each obstacle only with its per-timestep worst-case reachable set, an MIQP trajectory optimizer with hard collision-avoidance constraints and a variable-elimination technique to reduce decision variables, and a formal safety analysis under explicit velocity-bound and bounded-estimation-error assumptions. Ablation studies report up to 7.4x optimization speedup from variable elimination and improved feasibility from STSFCs; benchmarks across static, forest, and dynamic scenarios claim highest success rates with zero constraint violations; perception-only and hardware UAV trials are also reported.","tokens_in":2073,"tokens_out":720,"duration_ms":52957,"significance":"If the formal safety analysis is correct and the benchmark comparisons are fair, the work offers a practical advance in trading off speed and guaranteed safety for UAVs in partially unknown dynamic settings. The STSFC construction (tighter than full-horizon inflation) and the variable-elimination speedup are concrete engineering contributions that could be adopted by other hard-constraint planners. The hardware demonstration of fully onboard planning strengthens the practical relevance.","major_comments":[{"comment":"Formal safety analysis: the collision-free guarantee is derived under the modeling assumptions that obstacle velocities are known and bounded and that estimation errors are bounded. The manuscript's simulations and hardware trials use obstacles whose motion exactly respects the same velocity bounds employed by the planner; no sensitivity analysis or trials with bound violations (e.g., brief accelerations exceeding the declared speed) are provided. Because the central safety claim for unknown environments rests on these assumptions, the lack of robustness evidence is load-bearing.","section":"Formal safety analysis"},{"comment":"STSFC generator and MIQP constraints: the reachable-set inflation is performed per time layer rather than over the full horizon, which improves feasibility but makes the hard polytope constraints valid only if the per-layer over-approximation contains every possible obstacle position at that instant. The manuscript does not quantify the tightness of the reachable-set over-approximation or show that the resulting polytopes remain valid under the stated estimation-error bounds; this directly affects the “no constraint violations” benchmark claim.","section":"STSFC generator"},{"comment":"Variable elimination in the MIQP: the technique is reported to yield up to 7.4x speedup while preserving feasibility. Without an explicit statement (or proof sketch) that the eliminated variables do not relax the hard collision-avoidance inequalities, it is unclear whether the reduced problem remains equivalent to the original MIQP; this is central to the claim that hard constraints are maintained at reduced cost.","section":"Trajectory optimization"}],"minor_comments":[{"comment":"The abstract states “six safe flights in static environments and ten safe flights among dynamic obstacles” but provides no quantitative metrics (e.g., minimum clearance, replanning frequency, or failure modes) for these trials.","section":"Abstract"},{"comment":"Figure captions and table headings should explicitly state whether the reported success rates and computation times are averaged over how many Monte-Carlo runs and whether error bars or standard deviations are shown.","section":"Benchmark results"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the thorough and constructive review. The comments help clarify the scope of our safety claims and strengthen the presentation of the technical contributions. We respond point-by-point to the major comments below.","responses":[{"response":"We agree that the collision-free guarantee in Section IV is conditional on the explicit assumptions of bounded obstacle velocities and bounded estimation errors. All reported simulations and hardware trials operate strictly inside these bounds, consistent with the standard 'unknown but bounded' modeling framework used in robust motion planning. We acknowledge that sensitivity analysis under deliberate bound violations is absent and would provide additional insight into practical robustness. In the revised manuscript we will expand the Limitations section to discuss the consequences of bound violations and to recommend the use of conservative velocity and error bounds in deployment. New empirical trials that intentionally violate the bounds lie outside the scope of the present work but are noted as valuable future validation.","revision_made":"partial","referee_comment":"Formal safety analysis: the collision-free guarantee is derived under the modeling assumptions that obstacle velocities are known and bounded and that estimation errors are bounded. The manuscript's simulations and hardware trials use obstacles whose motion exactly respects the same velocity bounds employed by the planner; no sensitivity analysis or trials with bound violations (e.g., brief accelerations exceeding the declared speed) are provided. Because the central safety claim for unknown environments rests on these assumptions, the lack of robustness evidence is load-bearing."},{"response":"The STSFC construction inflates each obstacle at every discrete time layer by its worst-case reachable set under the velocity bound and then further enlarges the polytope by the estimation-error radius ε. Under the paper's assumptions this guarantees containment of every possible obstacle position at that instant. To address the request for explicit quantification, the revised manuscript will add a short lemma in Section III-B that (i) states the containment property formally and (ii) provides a simple volume-based bound on the over-approximation tightness for the chosen reachable-set model. This addition will directly support the validity of the hard polytope constraints and the reported zero-violation results.","revision_made":"yes","referee_comment":"STSFC generator and MIQP constraints: the reachable-set inflation is performed per time layer rather than over the full horizon, which improves feasibility but makes the hard polytope constraints valid only if the per-layer over-approximation contains every possible obstacle position at that instant. The manuscript does not quantify the tightness of the reachable-set over-approximation or show that the resulting polytopes remain valid under the stated estimation-error bounds; this directly affects the “no constraint violations” benchmark claim."},{"response":"The variable-elimination procedure substitutes only those decision variables that are affine functions of the retained variables (primarily positions expressed via integrated velocity/acceleration). Because the collision-avoidance inequalities are linear in the position variables, the substitution preserves the feasible set exactly; no relaxation is introduced. The revised manuscript will include a concise proof sketch in the appendix demonstrating equivalence of the reduced and original MIQPs with respect to both feasibility and the optimal objective value. This will confirm that the reported speed-up is achieved without weakening the hard constraints.","revision_made":"yes","referee_comment":"Variable elimination in the MIQP: the technique is reported to yield up to 7.4x speedup while preserving feasibility. Without an explicit statement (or proof sketch) that the eliminated variables do not relax the hard collision-avoidance inequalities, it is unclear whether the reduced problem remains equivalent to the original MIQP; this is central to the claim that hard constraints are maintained at reduced cost."}],"tokens_in":1666,"tokens_out":769,"duration_ms":54045,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper introduces a planner that first runs a heat-map A* to steer clear of high-risk zones, then generates spatiotemporal safe flight corridors by expanding each obstacle only with its worst-case reachable set at that specific time layer rather than the full horizon. This produces tighter corridors that stay feasible in dense settings. The trajectory is optimized via MIQP with hard collision constraints, but a variable-elimination step cuts the decision variables and yields up to 7.4x faster solves. A formal safety argument is given under bounded obstacle velocities and bounded estimation error. Benchmarks on static forests, dynamic scenes, and perception-only runs show higher success rates than prior methods with zero reported violations, and onboard UAV flights succeeded in both static and moving-obstacle cases. The engineering choices around per-layer inflation and the MIQP reduction are concrete and address the speed-safety trade-off directly. Ablations confirm that both the STSFC and the variable elimination matter for performance in cluttered dynamic environments. The soft spot is that the collision-free guarantee and the hard constraints hold only when the declared velocity bounds are never exceeded and the reachable-set over-approximation remains valid. The reported zero-violation results come from simulations and hardware trials where obstacles obey exactly those bounds; in genuinely unknown settings an unmodeled acceleration or sensing spike could invalidate the corridors without the planner noticing. The abstract states the assumptions clearly, so the limitation is not hidden, but reviewers will want to see how tight the over-approximations actually are and whether the proof accounts for discretization effects. This work is aimed at robotics groups building real-time 3D planners for UAVs and mobile robots that must operate without prior maps. It is worth sending to peer review because the techniques are specific, the experiments include both standardized benchmarks and hardware, and the formal analysis is present even if its tightness needs checking.","headline":"SANDO's main contribution is the time-layered STSFC that inflates obstacles only by their per-step reachable set plus variable elimination in the MIQP, which delivers faster feasible plans than prior hard-constraint methods while keeping formal safety under explicit velocity bounds.","tokens_in":2575,"tokens_out":462,"would_cite":false,"duration_ms":33466,"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":"SANDO generates collision-free trajectories in unknown dynamic 3D environments by steering paths with risk heat maps and enforcing hard avoidance via time-specific obstacle corridors in a reduced MIQP.","keywords":["trajectory planning","dynamic unknown environments","safe flight corridors","MIQP optimization","UAV navigation","collision avoidance","formal safety analysis","spatiotemporal planning"],"falsifier":"A recorded collision occurring when an obstacle's actual velocity exceeds the bound used to compute its reachable set in the corridors, despite the planner completing without violations.","tokens_in":2738,"feed_emoji":"🚁","tokens_out":766,"duration_ms":41432,"temperature":0.7,"pith_summary":"The paper introduces SANDO to plan safe paths for UAVs when obstacle locations and velocities are unknown ahead of time and can change at any moment. A heat-map A* global planner first avoids high-risk areas using soft costs, then a generator builds spatiotemporal safe flight corridors as time-layered polytopes that expand each obstacle only by its worst-case reachable set at that exact time layer. Trajectory optimization solves a mixed-integer quadratic program with hard collision constraints, sped up by eliminating variables, while formal analysis proves no collisions under explicit bounds on obstacle speeds and sensing errors. Ablation tests confirm the corridors are essential for feasibility in dense settings and that variable elimination yields major speed gains. Benchmarks and onboard flights show higher success rates than prior methods with zero violations even when using only perception data.","feed_headline":"SANDO keeps UAVs collision-free among unknown moving obstacles","feed_subtitle":"Time-layered corridors and variable-reduced MIQP deliver top success rates plus formal safety in 3D dynamic tests.","key_machinery":"The spatiotemporal safe flight corridor (STSFC) generator, which builds time-layered polytopes using per-time worst-case reachable sets of obstacles to supply hard constraints to an MIQP solver whose variable count is reduced by elimination.","core_discovery":"SANDO produces safe trajectories by first using a heat map-based A* to guide the global path away from high-risk regions, then generating spatiotemporal safe flight corridors consisting of time-layered polytopes that inflate obstacles solely by their reachable sets at each time step rather than over the full horizon, and finally optimizing the trajectory as an MIQP that incorporates hard collision-avoidance constraints with a variable-elimination technique to enable rapid solving, all supported by formal safety guarantees under bounded-velocity and bounded-error assumptions.","pith_inferences":["Separating soft risk costs in global planning from hard constraints in local optimization may resolve the classic speed-safety trade-off for replanning tasks.","The time-layer approach to corridor inflation could reduce unnecessary conservatism compared with horizon-wide inflation in other dynamic planners.","Similar corridor and elimination techniques might transfer to ground robots or multi-agent coordination facing bounded but uncertain motions.","Tighter integration with learned obstacle predictors could relax the velocity-bound assumption while preserving the formal guarantees."],"forward_implications":["SANDO achieves the highest success rate with no constraint violations across static, forest, and dynamic benchmarks at all difficulty levels.","Variable elimination reduces optimization time by up to 7.4 times while STSFCs maintain feasibility in dense dynamic environments.","Perception-only operation without ground-truth obstacle data still delivers robust performance under realistic sensing.","Formal analysis guarantees collision-free paths given explicit velocity bounds and estimation-error limits.","Hardware tests confirm ten safe flights among dynamic obstacles with fully onboard planning, perception, and localization."],"fun_headline_variants":["SANDO secures collision-free paths for UAVs in unknown dynamic environments","SANDO steers paths using heat maps away from high-risk dynamic zones","Time-layered polytopes support SANDO safety in unknown obstacle motions","Variable elimination accelerates SANDO MIQP solving while preserving safety"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"Obstacles move with unknown but bounded velocities whose worst-case reachable sets can be computed accurately at each time layer without introducing hidden collisions or infeasibility.","fun_headline_variants_meta":{"raw":{"variants":["SANDO secures collision-free paths for UAVs in unknown dynamic environments","SANDO steers paths using heat maps away from high-risk dynamic zones","Time-layered polytopes support SANDO safety in unknown obstacle motions","Variable elimination accelerates SANDO MIQP solving while preserving safety"]},"model":"grok-4.3","cost_usd":0.011946,"raw_usage":{"total_tokens":5268,"prompt_tokens":764,"num_sources_used":0,"completion_tokens":70,"cost_in_usd_ticks":119462000,"prompt_tokens_details":{"text_tokens":764,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4434,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":764,"tokens_out":70,"duration_ms":83603,"temperature":1.0,"reasoning_tokens":4434,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-10T16:57:09.178741+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A recorded collision occurring when an obstacle's actual velocity exceeds the bound used to compute its reachable set in the corridors, despite the planner completing without violations.","supporting_citations":[],"review_version":1}