University of Twente Student Theses


Design and Optimisation of Roller Coaster Elements using Reinforcement Learning

Schuttert, W.J. (2022) Design and Optimisation of Roller Coaster Elements using Reinforcement Learning.

[img] PDF
Abstract:The design of roller coasters in a layout phase can be done traditionally by use of geometric and force vector designs. However, these conventional approaches are often characterised by a lack of flexibility, the need for expert knowledge and difficulty in integrating detailed design. In this research, a novel view on the design problem is taken by the help of which the roller coaster track design is seen as a reinforcement learning (RL) example. The design of a few specific track elements such as a drop, a hill, a banked turn, a 2D looping and a pitched turn is thus mathematically modelled as a Markov Decision Process (MDP), the core of the RL algorithm. The track element is modelled as a parameterized spline, whereas the physics of the roller coasters is integrated in two different ways: either as a simplified point mass or as a more detailed multi-body model. The parameters of the track are then found by optimizing rewards, here to be understood as design goals, that promote for mechanical stability, thrill and passenger safety. The optimal design is then achieved with the help of the proximal policy optimisation RL algorithm. Although the proposed automatic solution can successfully design some of the most important track elements, the RL solution is shown to struggle with the spatial awareness due to made simplified assumptions. Despite these shortcomings, the automatic RL design is perceived as a promising tool for the roller coaster application. The further focus thus can be on the achievement of the mechanical feasibility and improvement of the MDP model in terms of predefined actions and rewards.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:52 mechanical engineering, 54 computer science
Programme:Mechanical Engineering MSc (60439)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page