Search

A novel method to determine probabilistic operational safety bound for rotary-wing unmanned aircraft systems (UAS) traffic management is proposed in this paper. The key idea is to combine a determi...

To evaluate the robustness gain of Bayesian neural networks on image classification tasks, we perform input perturbations, and adversarial attacks to the state-of-the-art Bayesian neural networks, with a benchmark CNN model as reference. The attacks are selected to simulate signal interference and cyberattacks towards CNN-based machine learning systems. The result shows that a Bayesian neural network achieves significantly higher robustness against adversarial attacks generated against a deterministic neural network model, without adversarial training. The Bayesian posterior can act as the safety precursor of ongoing malicious activities. Furthermore, we show that the stochastic classifier after the deterministic CNN extractor has sufficient robustness enhancement rather than a stochastic feature extractor before the stochastic classifier. This advises on utilizing stochastic layers in building decision-making pipelines within a safety-critical domain.

Obstacle avoidance for small unmanned aircraft is vital for the safety of future urban air mobility (UAM) and Unmanned Aircraft System (UAS) Traffic Management (UTM). There are many techniques for real-time robust drone guidance, but many of them solve in discretized airspace and control, which would require an additional path smoothing step to provide flexible commands for UAS. To provide a safe and efficient computational guidance of operations for unmanned aircraft, we explore the use of a deep reinforcement learning algorithm based on Proximal Policy Optimization (PPO) to guide autonomous UAS to their destinations while avoiding obstacles through continuous control. The proposed scenario state representation and reward function can map the continuous state space to continuous control for both heading angle and speed. To verify the performance of the proposed learning framework, we conducted numerical experiments with static and moving obstacles. Uncertainties associated with the environments and safety operation bounds are investigated in detail. Results show that the proposed model can provide accurate and robust guidance and resolve conflict with a success rate of over 99%.

This paper proposed a novel method to determine probabilistic operational safety bound for unmanned aircraft traffic management. The key idea is to implement probabilistic uncertainty quantification and design the operational safety bound shape considering UAV’s heading direction. Operational safety bound is used to identify a virtual geographic boundary to protect aircraft and to ensure airspace safety. The proposed operational safety bound is calculated as a function of vehicle performance characteristics, state of vehicle, weather and other probabilistic parameters that affect the real position of vehicle such as position error from the Global Positioning System (GPS). It is calculated individually for each vehicle using real-time data and probability simulation. It considers the heading direction of vehicle and thus it is an anisotropic design. Monte Carlo simulations are conducted to estimate the operational safety bound size with a specified probability of failure. Results indicate that uncertainty is crucial for the operational safety bound’s size. Sensitivity study shows that UAV speed has the largest effect on the operational safety bound size. Analysis of impact of failure probability shows that operational safety bound size increases with the decrease in allowable failure probability, but the bound size based on different operational safety bound concept increases at different rate.

A probabilistic approach to help future UAS safety and traffic management is proposed. A probabilistic risk-based operational safety bound is first proposed. This will allow UAV to maintain the deviation from the trajectory plan within a ”buffer” zone, and keep away from obstacles. Uncertainty quantification, Monte Carlo simulation, and coordinate transformation techniques will be used. Additionally, an algorithm integrating reinforcement learning and the proposed operational safety bound will be developed which is used for collision avoidance and trajectory planning. This will provide a learning ability to UAVs in order to adapt their behavior to changing environments. Also, large-scale UAV management will be studied by using multi-agent generative adversarial imitation learning.