Robotics with GPU computing

Cupoch is a library that implements rapid 3D data processing and robotics computation using CUDA.

The goal of this library is to process the input of 3D sensors rapidly and use it to control the robot. This library is based on the functionality of Open3D, with additional features for integration into robot systems.

Core Features

3D data processing and robotics computation using CUDA
- Point cloud registration (ICP, ColoredICP)
- Point cloud clustering
  - G-DBSCAN: A GPU Accelerated Algorithm for Density-based Clustering
- Point cloud/Triangle mesh filtering, down sampling
- Visual Odometry
  - Real-time visual odometry from dense RGB-D images
  - Robust Odometry Estimation for RGB-D Cameras
- Collision checking
- Occupancy grid
- Distance transform
  - Parallel Banding Algorithm to Compute Exact Distance Transform with the GPU
- Path finding on graph structure
- Path planning for collision avoidance
Open3D-like API
Support memory pool and managed allocators
Interactive GUI (OpenGL CUDA interop and imgui)
Interoperability between cupoch 3D data and DLPack(Pytorch, Cupy,...) data structure

Installation

This software is tested under 64 Bit Ubuntu Linux 18.04 and CUDA 10.0/10.1/10.2. You can install cupoch using pip.

pip install cupoch

Or install cupoch from source.

git clone https://github.com/neka-nat/cupoch.git --recurse
cd cupoch
mkdir build
cd build
cmake ..; make install-pip-package -j

Installation for Jetson

You can also install cupoch using pip on Jetson. Please set up Jetson using jetcard and install some packages with apt.

sudo apt-get install libxinerama-dev libxcursor-dev libglu1-mesa-dev
pip3 install https://github.com/neka-nat/cupoch/releases/download/v0.0.9/cupoch-0.0.9.0-cp36-cp36m-linux_aarch64.whl

Results

The figure shows Cupoch's point cloud algorithms speedup over Open3D. The environment tested on has the following specs:

Intel Core i7-7700HQ CPU
Nvidia GTX1070 GPU
OMP_NUM_THREAD=1

You can get the result by running the example script in your environment.

cd examples/python/basic
python benchmarks.py

Visual odometry with intel realsense D435

Occupancy grid with intel realsense D435

Path finding

Visual odometry with ROS + D435

This demo works in the following environment.

ROS melodic
Python2.7

# Launch roscore and rviz in the other terminals.
cd examples/python/ros
python realsense_rgbd_odometry_node.py

Visualization

Point Cloud	Triangle Mesh	Voxel Grid

Occupancy Grid	Graph	Image

References

CUDA repository forked from Open3D, https://github.com/theNded/Open3D
GPU computing in Robotics, https://github.com/JanuszBedkowski/gpu_computing_in_robotics
Voxel collision comupation for robotics, https://github.com/fzi-forschungszentrum-informatik/gpu-voxels

baoyufuyou / cupoch