Details

Autor(en) / Beteiligte

• Wei, Zhenyu
• Shao, Zhijiang

Titel

Precision landing of autonomous parafoil system via deep reinforcement learning

Ist Teil von

• 2024 IEEE Aerospace Conference, 2024, p.1-10

Ort / Verlag

IEEE

Erscheinungsjahr

2024

Quelle

IEEE/IET Electronic Library (IEL)

Beschreibungen/Notizen

• This paper focuses on utilizing deep reinforcement learning (DRL) to achieve the precision landing of an autonomous parafoil system. Specifically, we present an integrated guidance and control algorithm realized by a policy network, which maps the state observation collected by the onboard sensor to the flap deflection control command. The reward function is designed to guide the policy network in minimizing both the touchdown error and the terminal anti-wind angle of the parafoil system. To ensure a smooth control profile during the time interval, an action space parameterization method based on Lagrange interpolation is proposed. On this basis, the policy network is trained by the proximal policy optimization algorithm, a type of on-policy DRL method. The parafoil dynamics are described using the six-degree-of-freedom model. Randomly distributed initial conditions are considered to account for the accumulated guidance error prior to terminal landing. Additionally, the Dryden wind disturbance is incorporated to account for the uncertainty of the wind field. The numerical results demonstrate that: i) The trained policy network effectively provides a smooth control profile for the autonomous parafoil system; ii) the policy network exhibits the potential to provide real-time command; iii) The integrated guidance and control algorithm based on DRL successfully enables precise terminal landing of the parafoil system, with a mean touchdown error smaller than 100m. The findings from this study will benefit the design of guidance and control algorithms for future autonomous parafoil systems employed in payload recovery missions or space exploration missions.