UAV Collision Avoidance Trajectory Planning With Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for mid-air collision avoidance in unmanned aerial vehicles (UAVs) are inadequate, particularly in congested airspace, as they rely on vertical separation and air traffic control, which can be ineffective in low-energy situations or with extended reaction times, necessitating a more efficient and autonomous collision avoidance system.
Innovation Solution
A reinforcement learning (RL) framework is employed to model aircraft dynamics and interactions, using a surrogate model to simulate maneuvers and generate a collision avoidance trajectory through a policy trained on a conservative surrogate task, ensuring feasibility and safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If vertical separation is used to prevent mid-air collisions, then collision avoidance is achieved through flight level assignment, but the system becomes ineffective in cases of air traffic congestion, low energy level, or extended reaction time
Solution Approach 1:
The patent transitions from traditional vertical separation (3D flight level assignment) to horizontal separation (2D surface navigation), fundamentally changing the dimension of collision avoidance. This allows UAVs to maintain safe distances laterally rather than vertically, improving adaptability in congested airspace where vertical separation becomes ineffective.
Solution Approach 2:
The system changes the separation parameter from vertical distance (flight levels) to horizontal distance (lateral separation distance). This parameter transformation enables the UAV to adapt to situations where vertical separation is unavailable or ineffective, such as in congested airspace or when energy constraints prevent climbing to different flight levels.
2Reliability
If air traffic control guidance is used for horizontal separation, then collision avoidance is provided through ATC rules, but the system requires extended reaction time and external control
Solution Approach 1:
The UAV implements autonomous collision avoidance by equipping it with onboard sensors, processors, and decision-making algorithms. The system independently detects obstacles, calculates safe trajectories, and executes avoidance maneuvers without requiring external ATC guidance, thereby eliminating reaction time delays associated with human operators and external control systems.
Solution Approach 2:
The system continuously monitors the environment and proactively identifies potential collision risks before they become imminent threats. By maintaining constant awareness and pre-calculating avoidance trajectories, the system reacts faster than traditional ATC-guided approaches that wait for conflict declaration before issuing resolution instructions.
3Reliability
If traditional collision avoidance systems are used, then basic safety is maintained, but the system cannot provide quick and feasible avoidance trajectories in complex airspace scenarios
Solution Approach 1:
The patent replaces traditional mechanical collision avoidance systems with an intelligent software-based reinforcement learning framework. This neural network-based system processes sensor data and generates avoidance trajectories through learned policies rather than predetermined mechanical rules, enabling rapid adaptation to complex airspace scenarios while maintaining safety constraints.
Solution Approach 2:
The system employs dynamic trajectory generation that adapts in real-time to changing environmental conditions, obstacle positions, and UAV states. Rather than relying on static avoidance rules, the reinforcement learning policy continuously adjusts the avoidance trajectory based on current situational awareness, enabling quick response to dynamic airspace conditions while guaranteeing safety constraints are met.
Data Source
AI summary
A method is provided for avoiding a conflict during a flight of an aircraft that includes a defined route of travel. The method includes receiving observations of states of the aircraft and a nearby obstacle in an environment of the aircraft as the aircraft travels the defined route. The method includes applying the states to a reinforcement learning framework to predict states of the aircraft to avoid a conflict between the aircraft and the nearby obstacle. The reinforcement learning framework determines maneuvers of the aircraft to avoid the conflict, using a policy trained using a surrogate model of the environment in which movements of the aircraft and the nearby obstacle are simulated, and determines the predicted states of the aircraft from the maneuvers. A collision avoidance trajectory is generated from the predicted states of the aircraft, and output for guidance, navigation or control of the aircraft.