elevation_mapping_cupy_core

A standalone, ROS-free Python package extracted from elevation_mapping_cupy (ETH Zurich / Takahiro Miki).

The upstream project is a full ROS2 node for real-time GPU-accelerated elevation mapping on legged robots. This package strips all ROS dependencies and exposes only the core Python library, making it usable in:

• Simulation pipelines (Isaac Sim, Gazebo, custom envs)
• Inference pipelines that consume elevation or semantic maps
• Any standalone Python code that needs GPU-backed elevation mapping without a ROS runtime

What's included

• ElevationMap — core CuPy-based elevation map with GPU kernels for point cloud fusion, traversability filtering, and map shifting
• SemanticMap — semantic layer management with pluggable fusion strategies (Bayesian, class-average, class-max, etc.)
• PluginManager — post-processing plugins (inpainting, erosion, smooth filter, traversability, PCA features, etc.)
• Parameter — unified configuration dataclass
• Custom CUDA kernels for fast point cloud ingestion and map updates

Installation

pip install elevation-mapping-cupy-core

Requires a CUDA 12.x environment with CuPy installed (cupy-cuda12x).

Quick start

from elevation_mapping_cupy import ElevationMap, Parameter

param = Parameter(resolution=0.04, map_length=8.0)
em = ElevationMap(param)

# Feed a point cloud (N, 3) numpy array
em.input(points, R, t, position_noise, orientation_noise)

# Retrieve the elevation layer as a numpy array
elevation = em.get_map_with_name("elevation")

License

MIT — Copyright (c) 2022 ETH Zurich, Takahiro Miki. See LICENSE.