OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

Contribution of an Artificial Intelligence approach to the Guidance/Piloting of PADS

Birolleau, Alexandre and Bedin, K. and Hangouet, Samuel and Ginestet, Jean-Brice and Perin, Basile and Galimant, R. and Celaudoux, V. and Pastor, Philippe and Patoux, Alice and Rivier, T. Contribution of an Artificial Intelligence approach to the Guidance/Piloting of PADS. (2022) In: 26th AIAA Aerodynamic Decelerator Systems Technology Conference, 16 May 2022 - 19 May 2022 (Toulouse, France).

[img] (Document in English)

PDF (Author's version) - Depositor and staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
3MB

Official URL: https://doi.org/10.2514/6.2022-2718

Abstract

The dropping of Precision Aerial Delivery Systems (PADS) aims at reaching the “higher, further and more reliable” triptych. Since the 2000s, the development of precise GPS and IMU position systems, and better estimation of wind and flight status have made possible more efficient control of PADS under a ram-air parachute. These allowed drops at ever-higher altitude and at ever further from target point. Nevertheless, the precision of the touchdown, involving the double issue of reaching a target point and the smoothness of the touchdown, has often shown shortcomings or at least a lack of reliability in the expected precision. Among the factors contributing to landing errors, unforeseen wind changes in the surface atmospheric layer are among the main causes of error. This is followed by inaccuracies in status estimation (sensors, real-time management, filtering), then by command modeling errors or also errors related to GNC algorithms. The necessary completeness and the proven complexity of mastering these aspects, generates development costs and significant experimental needs for a result that is not necessarily up to the level of the work. To overcome these limits of the "physical" (φ) or classic approach, a study was launched within the French Armament Procurement Agency DGA associated with its affiliated school ISAE-SUPAERO, to assess the possible relevance of applying Artificial Intelligence to manage the guidance and control of a load under a ram-air parachute, therefore by a so-called “AI” approach, and hence to improve the precision of PADS. Thus, the work reported here assesses the contribution of Reinforcement Learning (RL) technologies to the guidance of PADS. To do this, we will present the learning platform used, as well as its coupling to a ram-air parachute flight simulator based on the 9DDL flight dynamics model developed by ONERA [02]. We present the chosen RL algorithm, the learning process and some stability improvements in order to anticipate the reality gap between simulation and reality. A complementary work is also done to evaluate the robustness of the approach as for example here to make the comparison between different models (3DOF to 9DOF). At the end, the obtained results are then compared in the simulator against a classical approach based on the φ laws. These prerequisites lay the foundations for an evaluation of the relevance of the results obtained with regard to a double cross validation using a real mini-PADS prototype (EOLE).

Item Type:Conference or Workshop Item (Paper)
Audience (conference):International conference proceedings
Uncontrolled Keywords:
Institution:Other partners > Direction Générale pour l'Armement - DGA (FRANCE)
Université de Toulouse > Institut Supérieur de l'Aéronautique et de l'Espace - ISAE-SUPAERO (FRANCE)
Laboratory name:
Statistics:download
Deposited On:04 Jan 2023 10:54

Repository Staff Only: item control page