|3d-model||walkable environment||multi-layered environment|
The Polygonal Environment to Environment with Layers (PEEL) project is a large part of my research. Its goal is to study and create algorithms for extracting walkable environments from arbitrary polygonal soup, and to split this walkable environment into a set of layers that are embeddable in two-dimensional space. This layered representation, also known as a multi-layered environment, can be used to generate navigation meshes, perform nearest-neighbour queries in 3d, analyse pedestrian safety for complex environments and many more things.
We are working on a pipeline that converts this polygonal soup, step-by-step, into a layered representation of the walkable environment. Together with R.M. Polak and J.L. Vermeulen, I have worked on extracting the walkable environment from a 3d-model. This process is subdivided into many smaller steps. Three of the most important steps are the slope filter, the vertical clearance filter and the gap filter. The slope filter removes geometry that is too steep for a humanoid to stand on. Locations where pedestrians cannot stand up straight are removed by the vertical clearance filter. Small modelling errors and 'holes' are detected and fixed by the gap filter.
Next, this walkable environment is split in a set of 2d layers. While doing this, it tries to minimize the locations where the different layers meet. The reason for this is that transferring between two different layers during a simulation can be an expensive step. Also, other algorithms sometimes have a runtime dependency on this number of connections.
Publications within this project
- On the topology of walkable environments
- Annotating traversable gaps in walkable environments
- Performing multicut on walkable environments
- Separating a walkable environment into layers
ECM Simulation Framework
The Explicit Corridor Map (ECM) simulation framework can be used for simulation large human crowds. It utilizes a novel data structure that can compactly encode the walkable space of multi-layered environments. It also allows for efficient clearance queries.
Within this project I have made some minor contributions. These contributions mainly focus on cross-platform compatibility and cross-compiler compatibility of the existing code base. I have also worked together with J.L. Vermeulen on identifying the most efficient nearest neighbour data structure for this framework. Besides that, some of the projects I work on rely on this framework to generate pedestrian trajectories.