With the rapid development of low-altitude economy and drone technology, Unmanned Aerial Vehicle systems face significant safety challenges. Statistics from 1,500 UAV enterprises reveal system failure rates as high as 10⁻³ per flight hour, where battery degradation and motor malfunctions are primary contributors. These vulnerabilities necessitate high-reliability emergency landing capabilities, particularly for multirotor drones operating over populated urban areas where uncontrolled descent poses severe risks to ground personnel. The ICAROUS framework developed by NASA provides foundational autonomy for routine operations but lacks specialized emergency response mechanisms. This research addresses this gap through a hierarchical trajectory optimization method that significantly enhances safety and efficiency during critical failure scenarios.
Autonomous Operation Framework for Unmanned Aerial Vehicles
ICAROUS (Integrated Configurable Algorithms for Reliable Operations of Unmanned Systems) is a modular software architecture enabling autonomous decision-making through core components:
| Autonomy Category | Level | Capability |
|---|---|---|
| Single-Vehicle | 1 | Remote Guidance |
| 2 | Real-Time Fault Diagnosis | |
| 3 | Fault Recovery & Flight Adaptation | |
| 4 | Onboard Route Replanning | |
| Multi-Vehicle | 1-3 | Collaborative Operations |
| Swarm | 1-3 | Distributed Strategic Objectives |
The emergency decision module augments ICAROUS by continuously monitoring UAV states via communication buses. When critical anomalies like power deterioration are detected, it executes:
- Remaining flight time estimation using power models
- Reachable area calculation
- Landing site selection
- Trajectory regeneration
The power consumption model for multirotor UAVs during emergency descent is:
$$P(v) = P_m \left(1 + \frac{3v^2}{U_{tip}^2}\right) + P_n \left(\sqrt{1 + \frac{v^4}{4v_0^4}} – \frac{v^2}{2v_0^2}\right)^{1/2} + \frac{1}{2}d_0\rho sAv^3$$
where $P_m = \frac{\delta}{8}\rho sA \Omega^3 R^3$ and $P_n = (1 + k)\frac{W^{3/2}}{\sqrt{2\rho A}}$ represent blade profile and induced power during hover, respectively.
Emergency Landing Trajectory Optimization
Landing Site Selection and Path Planning
Candidate landing sites $S_l = \{l_1, l_2, …, l_n\}$ are predefined based on terrain features, population density, and obstacle distribution. Each site $l_i$ is characterized by:
$$l_i = \{ \mathbf{c}_i, r_{l,i}, r_{p,i} \}$$
where $\mathbf{c}_i \in \mathbb{R}^3$ denotes centroid coordinates, $r_{l,i} \in [0,1]$ indicates location risk, and $r_{p,i}$ represents path-dependent risk. Sites are filtered by reachability constraints:
$$d_i \leq d_{max} = \frac{E_t \cdot v}{P(v)}$$
Path planning integrates distance, climb penalties, and geofencing risks through an enhanced A* algorithm. The nodal transition cost is:
$$c(n,n_0) = \mu \cdot \text{dist}(n,n_0) + \varphi \cdot \text{climb}(n,n_0) + \lambda \cdot \text{risk}(n)$$
with cumulative cost $g(n) = g(n_0) + c(n,n_0)$ and heuristic $h(n) = \text{dist}(n,\text{goal})$. Risk assessment uses geofencing data:
$$\text{risk}(i,j,k) = \max\left(1 – \frac{D_{sta}(i,j,k)}{d_{sta}^{hold}}, 0\right)$$
The optimization objective selects the minimal integrated cost path:
$$\min_X J(X) = \min \left[ \sum_{j=1}^k \left( \mu \cdot \text{dist}(n_j,n_{j-1}) + \varphi \cdot \text{climb}(n_j,n_{j-1}) + \lambda \cdot \text{risk}(n_j) \right) \right]$$
Velocity Profile Optimization
Speed optimization is formulated as a Markov Decision Process with state space:
$$\mathbf{S}_t = [i_t, v_t, v_{h,t}, v_{v,t}]^T$$
Action space $\mathcal{A} = \{a_{\min}, a_{\min} + \Delta a, …, a_{\max}\}$ with $\Delta a = \frac{a_{\max} – a_{\min}}{N_a}$, and deterministic state transition:
$$\mathbf{S}_{t+1} = [i_t + 1, v_t + a_t \cdot \Delta t, v_{h,t+1}, v_{v,t+1}]^T$$
The reward function minimizes energy consumption while penalizing constraint violations:
$$R(\mathbf{S}_t, a_t) = -\left[ \frac{P(v_t)}{v_t} \cdot \Delta s + K \cdot \mathbb{I}_{\text{violation}} \right]$$
Value iteration solves the optimization:
$$\pi^* = \arg \min_\pi \sum_{t=1}^M \frac{P(v_t)}{v_t} \Delta s$$
Key constraints include:
$$0 \leq v \leq v_{\max}, \quad |a| \leq a_{\max}, \quad \int_0^L \frac{ds}{v(s)} \leq T_{\max}$$
Simulation Analysis
Path Planning Performance
Three urban scenarios (1,000m × 1,000m × 200m) demonstrate the algorithm’s effectiveness:
| Metric | Proposed Method | ICAROUS Baseline | Improvement |
|---|---|---|---|
| Cumulative Risk | 62.27 | 87.43 | -28.8% |
| Path Cost | 110.00 | 108.00 | +1.8% |
| Climb Penalty | 0.00 | 0.00 | 0% |
Scenario 1 illustrates critical risk-distance tradeoffs where the proposed method selects farther but safer landing sites despite 1.8% longer paths. Scenario 2 shows optimal conditions with multiple low-risk sites, while Scenario 3 demonstrates forced high-risk operations when alternatives are unavailable.
Velocity Optimization Results
Comparative analysis of velocity optimization methods:
| Algorithm | Energy (J) | Time (s) | Speed Fluctuation (m/s) |
|---|---|---|---|
| MDP (Proposed) | 7,261.05 | 47.45 | 0.253 |
| SQP | 7,609.81 | 53.16 | 0.719 |
| PSO | 8,491.32 | 64.80 | 1.047 |
The MDP approach generates a three-phase velocity profile: rapid ascent (≤3m/s² acceleration), steady cruise (15m/s), and smooth descent (≥-3m/s² deceleration). This reduces flight time by 10.7% and 26.8% compared to SQP and PSO, respectively, while decreasing speed fluctuations by 64.8% and 75.8%. Required Time of Arrival (RTA) allocation demonstrates temporal feasibility across trajectory waypoints.
Conclusion
This research presents a hierarchical trajectory optimization framework for multirotor UAV emergency landing under the ICAROUS architecture. The high-level planner integrates risk-aware A* pathfinding with Bézier curve smoothing, reducing cumulative risk exposure by 28.8% compared to native ICAROUS navigation. The low-level Markov Decision Process optimizes velocity profiles, achieving 10.7-26.8% time reduction and over 60% lower speed fluctuations versus conventional methods. Future work will address environmental disturbances and dynamic obstacle avoidance to enhance real-world applicability. This approach significantly advances drone technology safety paradigms for urban air mobility operations.