Efficient Global Navigational Planning in 3D Structures based on Point Cloud Tomography

TL;DR

This work proposes a highly efficient and extensible global navigation framework based on a tomographic understanding of the environment to navigate ground robots in multilayer structures that reduces the scene evaluation time by three orders of magnitude and improves the path planning speed by three times compared with existing approaches.

Abstract

Navigation in complex 3D scenarios requires appropriate environment representation for efficient scene understanding and trajectory generation. We propose a highly efficient and extensible global navigation framework based on a tomographic understanding of the environment to navigate ground robots in multi-layer structures. Our approach generates tomogram slices using the point cloud map to encode the geometric structure as ground and ceiling elevations. Then it evaluates the scene traversability considering the robot's motion capabilities. Both the tomogram construction and the scene evaluation are accelerated through parallel computation. Our approach further alleviates the trajectory generation complexity compared with planning in 3D spaces directly. It generates 3D trajectories by searching through multiple tomogram slices and separately adjusts the robot height to avoid overhangs. We evaluate our framework in various simulation scenarios and further test it in the real world on a quadrupedal robot. Our approach reduces the scene evaluation time by 3 orders of magnitude and improves the path planning speed by 3 times compared with existing approaches, demonstrating highly efficient global navigation in various complex 3D environments. The code is available at: https://github.com/byangw/PCT_planner.

Figures (9)

• Figure 1: We present a highly efficient and extensible global navigation framework that generates smooth 3D trajectories in complex multi-layer scenarios. It adopts a novel scene representation which enables rapid scene evaluation and further alleviates the burden of path searching. The framework applies to a wide range of ground robots including wheeled or quadrupedal robots and supports active height adaptation of the platform to avoid overhangs.
• Figure 2: We construct tomogram slices by projecting the point cloud onto a series of equidistant horizontal planes. Each slice contains a ceiling (orange) and a ground (blue) layer. The squares present the grid traversability (green: traversable, red: untraversable) considering the ground conditions and the ceiling height. As all the traversable grids in slice 2 are contained in the union of slices 1 and 3, slice 2 can be omitted as done in Section \ref{['sec:simp']}. The red-circled grids are the "gateways" defined in Section \ref{['sec:planning']} that connect slice 1 to the upper slices for searching upwards, as these grids share the same spatial position but the cost in the upper slice is lower and reflects the real traversability at that position. By using the gateways, our planner travels through multiple slices to search on the slope and under the overhangs, thus enabling navigation in multi-layer structures.
• Figure 3: (a): Illustration of the inflation kernel when $d_{inf}=0.2$, $d_{sm}=0.4$ and $r_g=0.1$. (b): The traversability cost map before inflation, where the untraversable grids are in orange and the traversable regions are in blue. The regions near the spiral center are untraversable due to the insufficient stair width. (c): The final travel cost map, where the untraversable regions are inflated and the costs are gradually reduced within the safe margin after applying the inflation kernel.
• Figure 4: (row a): The point cloud map (a1) of the spiral environment in beyond2D is used to construct our tomogram. The environment is then represented by five 2.5D tomogram slices (a2) to (a6) after simplification, where the gray layers are ceilings and the colored layers are ground elevations. (row b): The whole planning space (visualized as a point cloud (b1), blue: traversable, orange: untraversable) can be obtained by integrating (b2) to (b6) which are 2.5D travel cost maps with the related ground elevations in (a2) to (a6). Note that (b1) is only for visualization and we directly adopt (b2) to (b6) for trajectory generation in Section \ref{['sec:planning']} and \ref{['sec:optimization']}. (row c): The planner searches on multiple tomogram slices along the green arrows and generates the 3D trajectory. It can enter the upper slices through the gateways in red and can also search downwards through the gateways in blue.
• Figure 5: The simulation scenarios and the trajectories generated by our approach. The first row presents the 3D models of environments: factory (a1), building (b1), forest (c1), and overpass (d1). The second row shows the point clouds of the corresponding scenarios and the third row presents the integrated travel cost maps generated by our approach. Our approach provides feasible and smooth trajectories in these complex 3D scenarios with various structured or irregular terrain features and overhanging objects.