Figures (16)  Tables (6)

    • Figure 1. 

      Single rocker-actuated landing leg.

    • Figure 2. 

      Movable legs (Nos. 1−2, black) and fixed leg (No. 3, red).

    • Figure 3. 

      Dynamic model solution for the landing system.

    • Figure 4. 

      Dynamic perception strategy for the landing model.

    • Figure 5. 

      Schematic diagram of the DTFSF method.

    • Figure 6. 

      Flowchart of the DTFSF method.

    • Figure 7. 

      Illustration of a randomized path (m = 20).

    • Figure 8. 

      Landing terrain types and the corresponding CSM: (a) Basic, (b) obstacle, and (c) extreme.

    • Figure 9. 

      Recognition accuracy at different step sizes.

    • Figure 10. 

      Simulation results: (a) H for Case 1; (b) $ \sigma $ for Case 1; (c) H for Case 2; (d) $ {\hat{\tau }}_{D} $ for Case 2; (e) H for Case 3; (f) $ {\hat{\tau }}_{D} $ for Case 3.

    • Figure 11. 

      Diagram of the adaptive hierarchical control framework.

    • Figure 12. 

      Basic finite state machine framework for an intelligent multirotor UAV.

    • Figure 13. 

      Diagram of the AHCF's time synchronization with task priority.

    • Figure 14. 

      Working principles and flowchart of the landing process.

    • Figure 15. 

      Landing procedure in a real environment.

    • Figure 16. 

      Data from the entire process from INIT to LANDED. (a) Flight; (b) perception; (c) power; (d) status.

    • Terrains Landing feasibility 10 Random CR0.95 (%)
      Basic Flat ○ 100.00, 97.89, 96.84, 94.74, 100.00,
      98.95, 100.00, 100.00, 100.00, 99.47
      Slope ○ 89.47, 91.05, 96.32, 82.63, 93.16,
      82.11, 95.26, 94.74, 82.63, 94.74
      Obstacle Discrete ○ 66.84, 77.89, 55.79, 76.84, 73.68,
      64.74, 60.00, 57.89, 60.53, 77.89
      Continuous ○ 61.58, 57.37, 71.58, 55.79, 50.53,
      68.95, 55.26, 50.53, 52.11, 64.74
      Extreme Urban debris × 11.05, 7.37, 4.74, 5.79, 4.74, 16.84,
      5.79, 10.53, 7.37, 8.42
      Planetary analog × 12.11, 12.68, 10.53, 12.63, 10.00,
      12.11, 12.11, 9.47, 12.63, 14.21

      Table 1. 

      Quantitative evaluation of the terrains' CR.

    • Metric Full batch DTFSF
      Computational complexity $ \text{O}({k}^{2}) $ $ \text{O}({m}^{2}) $
      Real-time performance Latency: $ {T}_{\text{delay}}=k\Delta t $ Bounded latency: $ {T}_{\text{delay}}=m\Delta t $
      Memory footprint Requires storage of an $ \text{O}({k}^{2}) $ similarity matrix and $ \text{O}(k) $ data points. Fixed storage: $ \text{O}({m}^{2}) $ matrix and $ \text{O}(m) $ data points.
      Robustness to noise Susceptible to historical noise caused by global averaging. Enhanced robustness via dynamic window updates and localized outlier rejection.
      Hardware compatibility Requires high-performance processors (e.g., Graphics Processing
      Unit [GPU]) for large $ k $      		 Executable on microembedded systems (e.g., ARM
      Cortex-M4) with constrained resources.

      Table 2. 

      Comparative analysis: Full batch vs. DTFSF.

    • Metric Units Proposed (1) SMC (2) PID-C (3) $ {p}_{12} $ value† $ {p}_{13} $ value†
      RMSE‡ 10−3 m 6.79, 0.162 8.31, 0.244 29.5, 1.52 <0.001* <0.001*
      IAE‡ 10−2 m·s 2.91, 0.138 3.90, 0.213 18.6, 1.94 <0.001* <0.001*
      ISE‡ 10−4 m2·s 4.61, 0.218 6.95, 0.380 87.2, 9.10 <0.001* <0.001*
      ITAE‡ 10−2 m·s2 2.89, 0.098 4.13, 0.253 42.7, 1.94 <0.001* <0.001*
      ITSE‡ 10−4 m2·s2 2.03, 0.0859 4.09, 0.311 76.5, 4.36 <0.001* <0.001*
      CE‡ 10−2 N·m·s 36.5, 1.10 35.7, 1.52 35.7, 0.917 0.998* >0.999*
      * Significant group difference (proposed vs. SMC and PID-C). † A battery of multiple univariate one-tailed t-tests (with Holm–Bonferroni correction) was conducted to compare the proposed controller with the two baseline controllers on all six performance indices (n = 50 trials per group, α = 0.01 with the total number of tests set to 10). ‡ RMSE is defined as $ \sqrt{\dfrac{1}{T}\int\nolimits_{0}^{T}{e}^{2}dt} $; IAE is defined as $ \int\nolimits_{0}^{T}\left| e\right| dt $; ISE is defined as $ \int\nolimits_{0}^{T}{e}^{2}dt $; ITAE is defined as $ \int\nolimits_{0}^{T}t\left| e\right| dt $; ITSE is defined as $ \int\nolimits_{0}^{T}t{e}^{2}dt $; CE is defined as $ \int\nolimits_{0}^{T}\left| {\tau }_{M}\right| dt $.

      Table 3. 

      Statistical comparison of landing performance metrics (Case 4).

    • Sensor Primary role Advantages Limitations
      IMU Attitude estimation High update rate, low latency Attitude drift, vibration sensitivity
      GPS Global positioning High accuracy, no drift, wide coverage Low update rate, signal occlusion
      Barometer Relative altitude Low cost, power-efficient Thermal drift, airflow disturbance
      IR/ultrasonic Low-altitude sensing A mm-level resolution without light Short range, rain/fog attenuation
      Stereo vision Terrain mapping SLAM capability, rich data High computation, lighting-dependent
      LiDAR Precision ranging Electromagnetic Interference (EMI)-resistant, cm-level accuracy, long range High power, rain/fog attenuation
      Force Detecting collisions Fast response, power efficient Affected by vibration

      Table 4. 

      Sensors potentially required for the landing phase

    • Stages Flight
      MCU      		 DTFSF Landing
      MCU      		 LiDAR IR Force
      Cruise (>3 m) 100 MHz 10 kHz 48 MHz 5 Hz OFF OFF
      Near ground (<3 m) 100 MHz 10 kHz 480 MHz 10 Hz 1 kHz 1 kHz

      Table 5. 

      MCU and sensor duty cycling.

    • Modules Frequency Cycle CPU
      usage      		 Optimization Hardware
      dependency
      DTFSF 10 Hz 2.4 ms 2.4% CMSIS-DSP ART accelerator
      + I-cache
      FOSO 100 Hz 320 μs 3.2% CMSIS-DSP FPU + DSP
      ASMC 1 kHz 140 μs 14% FPU-accelerated Double-precision
      FPU
      PID-C 1 kHz 12 μs 1.2% − −
      Covering the key modules and the system's components. Time data were measured using Keil Microcontroller Development Kit (MDK) optimization compilation (-O3) and the on-board Discrete Wavelet Transformation (DWT) counter.

      Table 6. 

      Analysis of the core algorithm's execution time