Rapid trajectory design in complex environments enabled by reinforcement learning and graph search strategies

• Published in: Acta astronautica 2020-06, Vol.171, p.172-195
• Main Authors: Das-Stuart, A., Howell, K.C., Folta, D.C.
• Format: Article
• Language: English
• Subjects: Accessibility; Combinatorial analysis; Design; Exploitation; Machine learning; Mapping; Optimization; Search process; Trajectories
• ISSN: 0094-5765; 1879-2030
• DOI: 10.1016/j.actaastro.2019.04.037

Designing trajectories in dynamically complex environments is challenging and easily becomes intractable. Recasting the problem may reduce the design time and offer global solutions by leveraging phase space mapping patterns available as accessible regions, and the application of search techniques from combinatorics. A computationally efficient search process produces potential trajectory concepts to meet unique design requirements over a broad range of mission types, including low-thrust scenarios. A successful framework is summarized in terms of four components: (i) Accessible regions — establishing reachable regions within the design space for a given thruster/engine capability: (ii) Database exploitation — discretization of well known dynamical structures to form a searchable 2D or 3D volume or map: (iii) Automated pathfinding — exploiting machine learning techniques to determine the transport sequence to deliver an efficient path: (iv) Convergence/optimization — once the transport sequence is determined as a globally efficient concept, it is optimized locally by traditional numerical strategies. •Design framework broadens trajectory combinatorial search capability.•Reinforcement learning, and graph search techniques produce trajectory initial guesses.•Characteristics to influence trajectory geometry and performance in the trade space.