Adaptive guidance and integrated navigation with reinforcement meta-learning

This paper proposes a novel adaptive guidance system developed using reinforcement meta-learning with a recurrent policy and value function approximator. The use of recurrent network layers allows the deployed policy to adapt in real time to environmental forces acting on the agent. We compare the p...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Acta astronautica Ročník 169; s. 180 - 190
Hlavní autoři: Gaudet, Brian, Linares, Richard, Furfaro, Roberto
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elmsford Elsevier Ltd 01.04.2020
Elsevier BV
Témata:
Adaptive systems
Asteroids
Doppler radar
Engine failure
Guidance
Guidance (motion)
Landing guidance
Learning
Lidar
Mars
Mars environment
Mars landing
Meta learning
Navigation
Radio altimeters
Reinforcement learning
Guidance
Reinforcement learning
Landing guidance
Meta learning
ISSN:0094-5765, 1879-2030
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:This paper proposes a novel adaptive guidance system developed using reinforcement meta-learning with a recurrent policy and value function approximator. The use of recurrent network layers allows the deployed policy to adapt in real time to environmental forces acting on the agent. We compare the performance of the DR/DV guidance law, an RL agent with a non-recurrent policy, and an RL agent with a recurrent policy in four challenging environments with unknown but highly variable dynamics. These tasks include a safe Mars landing with random engine failure and a landing on an asteroid with unknown environmental dynamics. We also demonstrate the ability of a RL meta-learning optimized policy to implement a guidance law using observations consisting of only Doppler radar altimeter readings in a Mars landing environment, and LIDAR altimeter readings in an asteroid landing environment thus integrating guidance and navigation. •Adaptive guidance using meta-reinforcement learning.•Learns to handle engine failures, variations in mass and environmental forces.•Approach learns closed-loop controller.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0094-5765
1879-2030
DOI:10.1016/j.actaastro.2020.01.007