Optimizing the flight path for a scouting Uncrewed Aerial Vehicle

Abstract: Post-disaster situations pose unique navigation challenges. One of those challenges is the unstructured nature of the environment, which makes it hard to layout paths for rescue vehicles. We propose the use of Uncrewed Aerial Vehicle (UAV) in such scenario to perform reconnaissance across the environment. To accomplish this, we propose an optimization-based approach to plan a path for the UAV at optimal height where the sensors of the UAV can cover the most area and collect data with minimum uncertainty.

• The paper introduces a two-stage optimization framework that decomposes the 3D path planning into 2D spatial routing and altitude optimization for effective UAV navigation.
• The methodology employs MATLAB's SQP via fmincon to incrementally compute collision-free, sensor-maximizing trajectories in complex occupancy grids.
• The study highlights trade-offs in weight tuning and obstacle modeling, offering insights for enhancing scalability and automation in UAV path planning.

Optimization-Based Flight Path Planning for Scouting UAVs in Post-Disaster Environments

Motivation and Context

The paper addresses autonomous UAV navigation for scouting tasks in post-disaster environments characterized by unstructured terrain and unreliable infrastructure. In these situations, UAVs are leveraged for reconnaissance, environmental data acquisition, and communications relay, mitigating risk to human operators. The challenge is to compute efficient, safe paths in 3D occupancy grids with variable sensor coverage and minimal uncertainty, considering obstacles and environment constraints.

Problem Formulation and Decomposition

A 3D occupancy grid model represents the environment, where each cell encodes the probability of occupancy by obstacles. The goal is to find a collision-free trajectory from a start to a target position that maximizes sensor coverage and minimizes uncertainty. Recognizing the computational intractability of direct optimization in high-dimensional space, the authors propose a two-stage decomposition:

1. 2D Path Planning: Project the 3D occupancy grid onto the XY plane, treating obstacles of increasing Z cross-section as maximum-size prisms. Inflate obstacle boundaries to account for UAV dimensions. The path is computed incrementally via weighted multi-objective optimization:
   • Objective: Minimize weighted sum of Euclidean distance to target and per-step occupancy likelihood.
   • Constraints: Maximum step size, environment boundaries, and state continuity.
   • Optimization: Employ MATLAB's fmincon with SQP, where each timestep considers the current location and incrementally extends the path via local optimization.
2. Optimal Scouting Height: For each planned 2D waypoint, determine the optimal altitude that balances data quality versus sensor coverage—enforcing smooth altitude transitions, start/end height constraints, and UAV physical flight limits.

This partitioned approach reduces the dimensionality of the optimization, isolating spatial path-finding from altitude adjustment.

Implementation Details

2D Path Planning

• Occupancy Grid Construction: The XY-projected occupancy matrix is generated from the 3D grid. Obstacles are expanded by a buffer equal to the UAV's radius.
• Cost Function: $$Z_a = W_1 \|P_t - T\|^2 + W_2 \mathrm{Occupancy}(P_t)$$, tuned such that $W_1$ increases near the target to emphasize goal convergence.
• Stepwise Progression: Each optimization step solves for the next position, recursively building a waypoint sequence.
• Constraints:
  • $|p_t - p_{t-1}| \leq C_s$ (step size)
  • $p_s = p_{t-1}$ (state initialization for each optimization)
  • Boundary checks to keep $p_t$ in map limits.

Altitude Optimization

• Cost Function: $Z_p = |z - h|$, where $h$ is the empirically determined optimal scouting height.
• Constraints: Maximum vertical delta per step $\Delta z \leq C_z$; start/end heights fixed; vertical bounds per UAV specs.

Solver and Parameter Tuning

• Sequential Quadratic Programming: MATLAB’s SQP is utilized for both spatial and altitude optimization problems, leveraging fmincon in recursive mode.
• Dynamic Weighting: Weight $W_1$ is adaptively increased as UAV approaches the target, but manual tuning is required due to differences in scale with the occupancy term.
• Intermediate Targets: For large environments, intermediate waypoints mitigate optimization drift into obstacles.

Results and Observed Trade-Offs

The approach is validated on synthetic random occupancy grids and MATLAB's canonical 3D city block grid:

• Computational Efficiency: The recursive optimization strategy enables path generation in complex environments but exhibits slowdowns near dense obstacles, highlighting the sensitivity of SQP to gradient irregularity caused by abrupt occupancy transitions.
• Weight Tuning: Manual tuning of objective weights is non-trivial, especially when occupancy and distance terms have disparate scales. Path quality is impacted by suboptimal weighting, sometimes causing the agent to select inefficient detours.
• Obstacle Handling: Modeling all obstacles as prismatic incurs conservatism, forfeiting usable free space, though it simplifies computation and collision checks.
• Constraint Management: Adding explicit obstacle constraints directly into the optimizer impedes performance, yielding negligible benefit over inflated boundaries.

Practical Implications and Limitations

Applicability

• Path generation for disaster reconnaissance: Immediate applicability for UAV deployment in search-and-rescue, hazard mapping, and rapid topology assessment.
• Sensor-Aware Trajectory Planning: Maintains flight at altitude maximizing sensor range and data integrity, essential in challenging RF/visual environments.

Limitations

• Manual weight tuning: The lack of automated hyperparameter selection remains a limitation for real-time deployment.
• Obstacle approximation: The prismatic abstraction sacrifices maneuvering flexibility, suggesting a need for improved 3D modeling and collision handling.
• Scalability: While stepwise SQP reduces computational burden, scalability to larger grids or fleets remains restricted by optimization overhead and constraint complexity.

Directions for Future Work

Theoretical and practical enhancements can include:

1. Automatic Weight Adaptation: Incorporate meta-optimization (e.g., grid search, reinforcement learning) to dynamically set objective weights based on local environment statistics.
2. Obstacle Representation: Move towards adaptive, non-prismatic point cloud/FoV-aware obstacles using voxel carving or mesh abstraction.
3. Hybrid Algorithms: Fuse sampling-based methods (e.g., RRT*, SB-SQB) with continuous optimization for initialization, then refine heights via quadratic programming to improve runtime and solution optimality.
4. Distributed Planning: Extend for multi-agent settings to support swarm-based scouting and collaborative mapping under communication constraints.

Conclusion

This work delivers a computational approach to flight path planning for scouting UAVs in complex post-disaster scenarios through problem decomposition, multi-objective incremental optimization, and altitude refinement. Although effective for current applications, improvements in weight tuning automation, obstacle modeling, and scalability are essential for broader adoption. Its methodological framework provides a solid foundation for future research into hybrid, adaptive, and distributed UAV path planning strategies.