BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Paulo Rosa (Deimos) DTSTART:20230427T160000Z DTEND:20230427T170000Z DTSTAMP:20260423T003252Z UID:MPML/105 DESCRIPTION:Title: D eep Reinforcement Learning based Integrated Guidance and Control for a Lau ncher Landing Problem\nby Paulo Rosa (Deimos) as part of Mathematics\, Physics and Machine Learning (IST\, Lisbon)\n\n\nAbstract\nDeep Reinforce ment Learning (Deep-RL) has received considerable attention in recent year s due to its ability to make an agent learn how to take optimal control ac tions\, given rich observation data via the maximization of a reward funct ion. Future space missions will need new on-board autonomy capabilities wi th increasingly complex requirements at the limits of the vehicle performa nce. This justifies the use of machine learning based techniques\, in part icular reinforcement learning in order to allow exploring the edge of the performance trade-off space. The guidance and control systems development for Reusable Launch Vehicles (RLV) can take advantage of reinforcement lea rning techniques for optimal adaption in the face of multi-objective requi rements and uncertain scenarios.\n\nIn AI4GNC - a project funded by the Eu ropean Space Agency (ESA)\, led by DEIMOS and participated by INESC-ID\, t he University of Lund\, and TASC - a Deep-RL algorithm was used to train a n actor-critic agent to simultaneously control the engine thrust magnitude and the two TVC gimbal angles to land a RLV in 6-DoF simulation. The desi gn followed an incremental approach\, progressively augmenting the number of degrees of freedom and introducing more complexity factors such as nonl inearity in models. Ultimately\, the full 6-DoF problem was addressed usin g a high fidelity simulator that includes a nonlinear actuator model and a realistic vehicle aerodynamic model. Starting from an initial vehicle sta te along a reentry trajectory\, the problem consists of precisely land the RLV while ensuring system requirements satisfaction\, such as saturation and rate limits in the actuation\, and aiming at fuel consumption optimali ty. The Deep Deterministic Policy Gradient (DDPG) algorithm was adopted as candidate strategy to allow the design of an integrated guidance and cont rol algorithm in continuous action and observation spaces.\n\nThe results obtained are very satisfactory in terms of landing accuracy and fuel consu mption. These results were also compared to a more classical and industria lly used solution\, due to its capability to yield satisfactory landing ac curacy and fuel consumption\, composed of a successive convexification gui dance and a PID controller tuned independently for the non-disturbed nomin al scenario. A reachability analysis was also performed to assess the stab ility and robustness of the closed-loop system composed by the integrated guidance and control NN\, trained for the 1-DoF scenario\, and the RLV dyn amics.\n\nTaking into account the fidelity of the benchmark adopted and th e results obtained\, this approach is deemed to have a significant potenti al for further developments and ultimately space industry applications\, s uch as In-Orbit Servicing (IOS) and Active Debris Removal (ADR)\, that als o require a high level of autonomy.\n LOCATION:https://researchseminars.org/talk/MPML/105/ END:VEVENT END:VCALENDAR