Artificial Intelligence Techniques for Spacecraft Attitude Control and Estimation

• Country: Netherlands
• Summary: Artificial Intelligence Techniques for Spacecraft Attitude Control and Estimation
• NLT Ref No: 97359312
• Deadline: 02 Apr 2024
• Financier: Other Funding Agencies
• Purchaser Ownership: Government
• Tender Value: Refer Document
• Notice Type: Tender
• Document Ref. No.: 1-12107
• Purchaser's Detail:
  Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details. Login to see full purchaser details.

  Login to see details
• Description:
• Tenders are invited for Artificial Intelligence Techniques for Spacecraft Attitude Control and Estimation. Open Date: 20/02/2024 12:02 CET Closing Date: 02/04/2024 13:00 CET Price Range: 500 KEURO The objective of this activity is to develop AI based techniques to assist the design and tuning of attitude control and estimation algorithms and demonstrate improved performance with respect to industry-standard control and estimation methods.Description:The key challenges of spacecraft attitude control and estimation for autonomous systems is to jointly ensure performance optimality, robustness and stability. In the context of spacecraft attitude control, the increasing demands in terms of performance combined with autonomy requirements calls for more advanced and complex control tasks for such systems, which shall be able to efficiently adapt in unknown or partially unknown environments with uncertain dynamics properties, or even cope to unexpected changes due to failures. The most recent research in Reinforcement Learning (RL) have shown promising results in that domain and would provide efficient augmentation to state-of-the-art control techniques by providing an augmented adaptive controller which converge to optimality.The increasing number of high accuracy pointing missions or COTS-based missions (limits in performance and/or continuous availability) calls for further improvements of the standard attitude estimation methods. State estimation in classical AOCS designs is typically implemented with well-established model-based filtering techniques. Recent research on learning based neural network (NN) estimation algorithms have shown promising results in terms of accuracy of the state estimation, while being able to generalise well across different motions dynamics, environments, sampling rates, system model uncertainties and non-linearities.The activity comprises the development of AI-based control and estimation algorithms for AOCS, specifically: Development of RL based controller design and tuning, using the most promising RL algorithms to assist and/or augment the AOCS controller definition, based on state-of-the-art techniques; the expected benefit / improvement in terms of performance and autonomy (offline vs online learning) is assessed. Development of a learning assistance for the estimation function design and tuning based on the most promising NN techniques, and demonstrate the expected benefits in terms of performance, robustness to sensors outage, adaptability to the different mission scenarios and operational constraints, on-board implementation.Potential benefit for autonomous on-board sensors calibration using NN shall also be investigated. Optimization - in terms of design, tuning and VV process - of the control and estimation algorithms when integrated together with the rest of the AOCS functional chain.The activity shall ensure that the developed solutions satisfy the reliability standards of safety critical space systems (reproducibility, robustness, computational efficiency, guaranteed convergence). Both control and estimation solutions are combined in two use cases, where a properly tuned controller and estimator of an existing mission shall be compared to the new developed algorithms using the AI assistance process. The aim is to demonstrate the extent of performance improvement, as well as to overall assess the induced impact on the AOCS overall development process.A formal VV approach for the selected algorithms shall be defined, and shall include validation in high-fidelity simulator environment of the overall performance and robustness of the combined control/estimation solution and with target processor-in-the-loop tests on a representative bench.The proposed tasks for this activity are: Review promising RL-based techniques and neural networks-based techniques for assisting the design of spacecraft attitude control and estimation. Identify target mission use cases. Define and implement the augmented functions using the assistance of RL-based control and NN based attitude estimation algorithms. Assess augmented controller and estimator performance and robustness and compare with nominal solutions in dedicated simulator framework. Verification and validation on functional avionics bench with Processor-In-the-Loop (PIL). Assess overall impact on AOCS development process of the use of AI-assistance for control and estimation design and tuning (offline / online cases). Identify potential opportunities for deployment on upcoming in-orbit demonstration missions and possibly start preparation.Depending on the selected use cases, the work of the activity can be adapted to better fit the matching between AI techniques developed and system uses.
• Documents:

 Tender Notice