Aircraft Autonomy Training With Randomized Opponent Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solutions for autonomous vehicle control in two-aircraft scenarios are limited by training with predetermined fixed values for aircraft parameters, which restrict the ability of machine learning algorithms to generalize and perform effectively against a variety of aircraft, leading to degraded performance when encountering unexpected parameter values.
Innovation Solution
The method involves training a machine learning algorithm by generating randomly-sampled values for aircraft parameters, such as maximum speed and weapon range, and using these values in a simulated environment to improve the algorithm's ability to operate across a broader range of aircraft types, while normalizing and concatenating these values to maintain performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If predetermined fixed values for aircraft parameters are used during training, then the machine learning algorithm can be trained efficiently with consistent data, but the algorithm's ability to generalize to diverse aircraft types is restricted
Solution Approach 1:
The patent applies parameter changes by systematically varying aircraft parameters (such as maximum speed, weapon range, and maneuverability) during the training process. Instead of using fixed predetermined values, the training system introduces randomized parameter variations to expose the machine learning algorithm to diverse aircraft configurations, thereby improving generalizability while maintaining training efficiency through structured parameter sampling.
2Adaptability or versatility
If randomly-sampled parameter values are introduced during training, then the algorithm's generalizability improves, but the training complexity and computational requirements increase
Solution Approach 1:
The patent applies dynamics by transitioning from static fixed parameter values to dynamic randomized parameter values during training. The system implements a parameter sampling mechanism that dynamically generates varied aircraft configurations throughout the training process, allowing the algorithm to adapt to changing conditions while managing complexity through efficient random sampling rather than exhaustive parameter exploration.
3Adaptability or versatility
If the machine learning algorithm is trained with a broader range of parameter values, then its performance in diverse scenarios improves, but the training time and computational resources required increase
Solution Approach 1:
The patent applies partial action by implementing a balanced parameter sampling strategy that introduces sufficient diversity to improve generalizability without exhaustively exploring all possible parameter combinations. The system samples from defined parameter ranges with appropriate distribution, providing just enough variation to enhance performance across diverse scenarios while avoiding the computational burden of complete parameter space exploration.
Data Source
AI summary
An example method for training a machine learning algorithm (MLA) to control a first aircraft in an environment that comprises the first aircraft and a second aircraft can involve: determining a first-aircraft action for the first aircraft to take within the environment; sending the first-aircraft action to a simulated environment; generating and sending to both the simulated environment and the MLA, randomly-sampled values for each of a set of parameters of the second aircraft different from predetermined fixed values for the set of parameters; receiving an observation of the simulated environment and a reward signal at the MLA, the observation including information about the simulated environment after the first aircraft has taken the first-aircraft action and the second aircraft has taken a second-aircraft action based on the randomly-sampled values; and updating the MLA based on the observation of the simulated environment, the reward signal, and the randomly-sampled values.