Abstract
Energy management is a critical aspect of risk assessment for Uncrewed Aerial Vehicle (UAV) flights, as a depleted battery during a flight brings almost guaranteed vehicle damage and a high risk of human injuries or property damage. Predicting the amount of energy a flight will consume is challenging as routing, weather, obstacles, and other factors affect the overall consumption. We develop a deep energy model for a UAV that uses Temporal Convolutional Networks to capture the time-varying features while incorporating static contextual information. Our energy model is trained on a real-world dataset and does not require segregating flights into regimes. We illustrate an improvement in power predictions by 29% on test flights when compared to a state-of-the-art analytical method. Using the energy model, we can predict the energy usage for a given trajectory and evaluate the risk of running out of battery during flight. We propose using Conditional Value-at-Risk (CVaR) as a metric for quantifying this risk. We show that CVaR captures the risk associated with worst-case energy consumption on a nominal path by transforming the output distribution of Monte Carlo forward simulations into a risk space. Computing the CVaR on the risk-space distribution provides a metric that can evaluate the overall risk of a flight before take-off. Our energy model and risk evaluation method can improve flight safety and evaluate the coverage area from a proposed takeoff location.
Introduction
For Uncrewed Aerial Vehicles (UAVs), battery life is a primary safety constraint. Running out of power mid-flight almost certainly leads to a crash, risking property damage or human injury. Predicting energy usage is notoriously difficult because it is a "moving target" influenced by changing wind speeds, air density, payload weight, and complex aerodynamic interactions. Furthermore, most existing risk assessments look at average energy use, which ignores the "long tail" of worst-case scenarios—like a sudden, sustained headwind that could drain a battery just short of the landing pad.
Methodology
The researchers developed a two-pronged approach to address these challenges: a deep learning-based energy model and a robust risk assessment metric.
Deep Energy Model
Instead of relying on simplified physics, the system uses Temporal Convolutional Networks (TCNs). TCNs are excellent at capturing time-varying features (like changes in airspeed and pitch) while remaining more memory-efficient and stable than traditional Recurrent Neural Networks. The model architecture separates inputs into time-variant data (airspeed, vertical speed, angle of attack) and time-invariant data (air density, payload mass) to predict power consumption at every second of a flight.
CVaR-based Risk Assessment
To handle environmental uncertainty, the team runs Monte Carlo simulations of planned flight paths under various wind conditions using CFD-modeled urban wind and Dryden turbulence. Instead of just looking at the average energy used, they calculate the Conditional Value-at-Risk (CVaR). CVaR focuses specifically on the "tail" of the distribution—the expected energy cost of the worst 5% of possible scenarios. This provides a safer "go/no-go" metric for flight operators compared to traditional mean-variance metrics.
Experimental Results
The model was trained and validated on a publicly available dataset of 209 flights using a DJI Matrice M100, totaling over 10 hours of real-world flying time with varying altitudes, speeds, and payloads.
Key results include:
- Accuracy Boost: The deep energy model showed a 29% improvement in power prediction over state-of-the-art analytical methods.
- Low Error: The best TCN model achieved a Mean Absolute Percentage Error (MAPE) of approximately 9% on unseen test flights.
- Efficiency: A "small" TCN (s-TCN) with 93% fewer parameters performed nearly as well as the larger version, making it a prime candidate for edge computing directly on drone hardware.
Conclusion
By combining TCN-based power prediction with CVaR risk metrics, this framework provides a robust way to quantify UAV battery risk under uncertainty. This methodology allows for the generation of risk contour maps to identify safe operational zones in complex urban environments. Future work will extend this to more aggressive flight maneuvers and different vehicle platforms like fixed-wing or VTOL craft.
Project Resources: * Video: https://youtu.be/PHXGigqilOA * Code: https://git.io/cvar-risk
Summary updated