Traditional path planning for unmanned aerial vehicles (UAVs) and autonomous underwater vehicles (AUVs) relies heavily on a simplifying assumption: operating at a fixed altitude or depth. By collapsing a three-dimensional world into a two-dimensional plane, optimization algorithms become computationally tractable. However, as highlighted by Peng Yang et al., this fixed-altitude constraint severely limits a vehicle's ability to navigate complex urban environments or exploit vertical spatial diversity.
In the real world, cities are dense with high-rise obstacles, and the ocean contains varying topographies. Restricting a vehicle to a single horizontal slice forces it to take elongated, inefficient detours around obstacles that it could otherwise simply fly over or dive under. Three-dimensional trajectory optimization removes these artificial geometric constraints, allowing vehicles to utilize the full volume of their environment.
The Non-Linear Physics of Vertical Motion
We cannot treat the vertical dimension as just another independent linear coordinate. The physics of propulsion introduces a fundamental asymmetry between horizontal and vertical movement. Moving horizontally requires overcoming aerodynamic or hydrodynamic drag at a relatively constant rate. Moving vertically, however, introduces the direct, heavy tax of gravity.
As Yuanhang Qi et al. point out, climbing, descending, and directional banking introduce highly non-linear energy costs. A short 2D path that requires aggressive altitude changes might consume significantly more energy than a longer, smoother 3D trajectory. Furthermore, Luis Rodrigues demonstrates that ignoring the energy dynamics of climb and descent phases misrepresents the true efficiency profile of an aircraft.
For underwater systems, the same principle applies in reverse; Niankai Yang et al. observe that omitting the energy costs of complex 3D diving and surfacing leads to inaccurate energy management models during deep-sea missions. True 3D trajectory optimization must therefore model these asymmetric physical constraints directly within the path-planning loop to ensure that vertical maneuvers actually save energy rather than waste it.
Maximizing Communication and Sensing Geometry
Beyond energy and obstacle avoidance, the vertical dimension dictates the quality of wireless communication and remote sensing. In aerial wireless networks, a UAV's altitude directly affects its Line-of-Sight (LoS) probability with ground users. If the drone flies too low, buildings block the signal; if it flies too high, path loss degrades the connection.
Xiao Liu et al. emphasize that modeling full three-dimensional maneuvers—including vertical displacement, velocity, and acceleration—is an open challenge essential for improving network endurance and signal quality. This becomes even more critical in high-precision applications like Free-Space Optical (FSO) communication.
Hyung-Joo Moon et al. show that integrating three-dimensional jitter models (accounting for roll, pitch, and yaw instabilities) into flight path planning yields substantial gains in energy efficiency and spectral performance. This proves that mechanical orientation in 3D space is deeply coupled with communication performance; a drone cannot maintain an optimal optical backhaul link without dynamically adjusting its altitude and attitude to counteract atmospheric and structural jitter.
Algorithmic Paradigms for 3D Control
To handle the compounding complexities of 3D environments, non-linear energy profiles, and dynamic channel states, researchers are shifting away from rigid geometric heuristics toward learning-based and predictive control frameworks. Multi-agent deep reinforcement learning (DRL) and economic model predictive control (EMPC) have emerged as dominant paradigms.
These methods allow vehicles to learn or predict the long-term consequences of vertical maneuvers, balancing the immediate energy cost of a climb against the long-term benefit of a clearer communication link or a shorter path. By optimizing trajectories across all three physical dimensions simultaneously, these systems unlock the full operational potential of autonomous mobile nodes.