ML Orbit Planner for Fast Spacecraft Transfer Trajectories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current trajectory planning for spacecraft using physics-based models is time and computationally intensive, hindering the generation of training data for machine learning and requiring significant bandwidth, which slows down the process of generating meaningful best guesses for high-fidelity orbit planning tools.
Innovation Solution
The implementation of a neural orbit planner or a gradient boosted tree (GBT) model to rapidly generate transfer orbits, providing medium-fidelity models that can be used as best guesses for high-fidelity modeling tools, significantly reducing the time needed for orbital planning and computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based models are used for trajectory planning, then high accuracy is achieved, but time consumption and computational bandwidth increase significantly
Solution Approach 1:
The system performs preliminary action by pre-computing training trajectories using physics-based models during an offline training phase. This pre-computed data is stored and later used to train machine learning models, which can then rapidly generate trajectory predictions without requiring real-time physics-based calculations, thus resolving the time consumption issue while maintaining accuracy through the trained model's predictions
Solution Approach 2:
The patent introduces machine learning models as an intermediary between the physics-based models and the real-time trajectory planning application. The ML models are trained on data from physics-based models and then serve as a faster approximation layer, mediating between the high-accuracy but slow physics-based models and the need for rapid real-time predictions, thereby reducing time consumption while preserving accuracy
2Measurement precision
If physics-based models are used for trajectory planning, then high accuracy is achieved, but computational bandwidth consumption increases
Solution Approach 1:
The system performs preliminary action by pre-computing training trajectories using physics-based models during an offline training phase. This pre-computed data is stored and later used to train machine learning models, which can then rapidly generate trajectory predictions without requiring real-time physics-based calculations, thus resolving the time consumption issue while maintaining accuracy through the trained model's predictions
Solution Approach 2:
The patent introduces machine learning models as an intermediary between the physics-based models and the real-time trajectory planning application. The ML models are trained on data from physics-based models and then serve as a faster approximation layer, mediating between the high-accuracy but slow physics-based models and the need for rapid real-time predictions, thereby reducing time consumption while preserving accuracy
3Measurement precision
If traditional methods are used for generating training data, then accurate orbital data is produced, but the process is time-consuming and computationally intensive
Solution Approach 1:
The system performs preliminary action by pre-computing training trajectories using physics-based models during an offline training phase. This pre-computed data is stored and later used to train machine learning models, which can then rapidly generate trajectory predictions without requiring real-time physics-based calculations, thus resolving the time consumption issue while maintaining accuracy through the trained model's predictions
Solution Approach 2:
The patent employs copying by using machine learning models to replicate and approximate the behavior of physics-based models. The ML models are trained to copy the trajectory generation capability of physics-based models but execute much faster, effectively creating a simplified copy that maintains accuracy while dramatically improving data generation speed and productivity
Data Source
AI summary
Discussed herein are devices, systems, and methods for improved trajectory planning. A method can include providing two of (i) a first value indicating a change in velocity to alter an orbit of a first object to a transfer orbit; (ii) a second value indicating a range between the first object and a second object; or (iii) a third value indicating an altitude of the first object relative to a celestial body around which the first and second objects are orbiting as input to a machine learning (ML) model, receiving, from the ML model, a holdout value, the holdout value a prediction of the value, of the first value, the second value, and the third value, that was not provided to the ML model, and providing the holdout value to an orbital planner.