KinoUAV: A Benchmark for Safe and Stable UAV Vision-Language Navigation with Kinodynamic Modeling
Existing UAV vision-language navigation benchmarks generate trajectories without kinodynamic constraints, leading to jerky, collision-prone flight. KinoUAV introduces a dataset built from acceleration-driven, kinodynamically feasible trajectories with directional free-space distance labels and a two-stage Cruise–Seek task. The proposed KinoUAV-Agent hierarchically plans waypoints and predicts continuous accelerations, reporting state-of-the-art success and stability on the benchmark.
The paper presents a well-motivated, internally consistent benchmark for safe UAV VLN. Its kinodynamic trajectory generation and DFD supervision are meaningful additions, and the hierarchical agent improves success and stability on the proposed benchmark. However, evaluation is limited to simulated scenes with simplified dynamics, and generalization to unseen maps falls sharply, tempering claims of robust aerial navigation.
The dataset construction and intermediate supervision are the strongest parts. The hierarchical kinodynamic A* pipeline explicitly enforces velocity, acceleration, and jerk limits during generation, producing more realistic training data than simple waypoint interpolation, and the ablation shows that adding the Auxiliary Subset improves success rate by 18.1 percentage points. The DFD branch cleanly decouples obstacle perception from action selection; removing it raises collision rate by 4.3 percentage points.
The evaluation relies on a small set of retro-fitted baselines and an as-yet-unvalidated benchmark. CMA, TravelUAV, and AerialVLA were adapted to KinoUAV's quantized acceleration action space—for example, AerialVLA's velocity commands were converted into three quantized components—so the 37-point success-rate gap in Table II is partly an apples-to-oranges comparison. The zero-shot VLM results (SR below 4%) act as a strawman because the task interface is clearly designed for the proposed fine-tuned pipeline. More importantly, success on unseen maps drops from 89.3% to 64.6%, suggesting that gains owe much to scene familiarity rather than robust kinodynamic reasoning.
The evidence supports internal claims that kinodynamic modeling and DFD labels improve KinoUAV metrics, but it does not establish superiority on prior aerial VLN benchmarks such as OpenFly or AerialVLN. The paper fairly notes that OpenFly automates collection without kinodynamic constraints—a claim corroborated by OpenFly's own description of using predefined flight actions as basic units—yet cross-benchmark experiments are absent. The introduction of absolute yaw rate as a stability metric is welcome, though reporting it on standardized prior benchmarks would allow direct comparison.
Experimental details are thorough—drag coefficients, cost weights, and LoRA rank are listed—but independent reproduction is blocked until the dataset and code are released. The GitHub link is promised but not yet populated at submission, and target descriptions are generated by the closed-weights Qwen3-VL-235B, making exact replication expensive. The 16 UE4 scenes come from OpenUAV, which aids reproducibility, though the voxel occupancy maps and exact target placements must also be shared.
Unmanned aerial vehicles (UAVs) have recently emerged as important remote sensing platforms, and UAV navigation has shown great potential in remote sensing applications. However, existing methods do not explicitly incorporate kinodynamic modeling or safety distances into trajectory generation, resulting in high absolute yaw rates, frequent collisions, and low navigation success rates, which limit safety and motion stability. To address this challenge, we introduce KinoUAV, a kinodynamic benchmark comprising a dataset and KinoUAV-Agent. First, we develop a hierarchical kinodynamic trajectory generation pipeline that drives UAV through acceleration actions. It uses waypoint chains from global planning and generates trajectories segment by segment under velocity, acceleration, and jerk constraints, reducing oscillations and improving trajectory stability. Furthermore, we introduce directional free-space distance (DFD) labels into the constructed dataset to describe the available clearance for safe motion decisions. The proposed KinoUAV-Agent predicts DFDs and uses them to generate safe actions, decoupling perception from action decision-making. This design enables autonomous obstacle avoidance and improves navigation safety. The benchmark incorporates the proposed Cruise-Seek task, in which the UAV first enters the target region and then searches for the target, reducing the difficulty of target search and improving navigation success. Experiments on KinoUAV show that KinoUAV-Agent outperforms existing methods, achieving the highest success rate while maintaining lower absolute yaw rates and collision rates. These results demonstrate the effectiveness of kinodynamic modeling for safe and stable UAV navigation.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.