LSMv2

LiDAR SiMulator v2: Interactive 2D LiDAR scanner simulator version II. Implemented via Rust and CUDA.

Functionalities & Features

• Simulated 2D scanner data. The simulation algorithm is written in CUDA (fast ray trace algorithm being able to simulated a scan with 2400 points within 0.5ms).
• Visualization via Rust nannou (implemented in OpenGL), enabling smooth and limitless visualization.
  • For example, the older simulation visualization implemented in OpenCV is restricted to pixel plotting: unable (or very tricky) to zoom in\out, rotate and translate the canvas.
  • Nannou has no limits w.r.t canvas operations and it is extremely efficient due to the bottom layer implementation of OpenGL.
  • nannou_egui crate used. Intergrated Rust EGUI GUI implementation, easy to use & good looking.

GUI for editor	GUI for simulator

• Smooth mouse & keyboard control of the simulated scanner (via nannou API).
• Mixed programing: Rust & CUDA C++: Enabling accelerated 2d LiDAR ray trace.
• Fast collision detection: scanner will not be allowed to drive into the obstacles
  • Coarse occupancy grid (indicates the obstacle id of the grid if the grid is occupied)
  • Use the id to access the contour of the obstacle and check whether the scanner is going to move inside the obstacle (point in polygon test)
  • An efficient algorithm implemented via Rust
• Better map editor (than version I, the old LiDARSim2D)
  • Canvas translation, rotation and scaling are supported.
  • Three modes: polygon, straight line (axis-aligned), rectangle (axis-aligned)
  • Drawn point selection, movement, deletion.
  • Easy to edit the previously drawn map. Trajectory display.
  • Screen shot API (also available in simulator)
• Rust implementation: Safer, clearer and efficiency guaranteed! (Personally, I prefer Rust than C++)
• Prettier front-end! Night mode supported!
• configuration parser: modify the configurations of LiDAR, map, etc. in a .json file without compiling!

TODO

• Dynamic obstacle simulation:
  • Trajectory and pose planning for one or multiple automatic mobile agents. (Like simulate pedestrians wandering in the map)
  • Dynamic obstacle ray trace (a few additions in my current CUDA ray tracer)
  • Interactable dynamic objects (doors, mainly)
• Automation & Editability
  • Trajectory recorder for semi-automatic simulation
  • Trajectory and pose planning for scanner --- fully automatic simulation
  • (On going) Better map editor and visualizer
• ROS API
  • Connect to ROS (I don't know how to do this, yet), send simulated sensor_msgs/LaserScan to rviz
  • More available sensors like Odometry and IMU
• Fancier Future
  • Network API: multiple scanners.
  • Visualizer. Replacing RViz!

Simple Demo

The following video is a demo for simulator:

1.mp4

The following video is a demo for the basic functionalities of LSMv2:

2.mp4

The following video is a demo for the different plotting mode of LSMv2:

3.mp4

Compile & Run

CUDA and rustc are of course, mandatory dependencies. arch=compute_86 is recommended, since its usability is verified. For lower versions, please modify lsm/build.rs:

...
.flag("-gencode")
.flag("arch=compute_86,code=sm_86")

Once you have the correct CUDA (and nvidia driver) and Rust Env, compilation should be smooth (if the source of Rust is correctly set). For (us) Chinese users, the source of rust libs should be changed to some faster domestic mirror sources, otherwise... wait for 10 years and the compilation might complete.

To run the simulator:

cd lsm
cargo build --release
cargo run --release --bin=sim

To run the editor:

cd lsm
cargo build --release
cargo run --release --bin=edit

And there u go, a screen pops up and after setting the initial pose (left button, long press), you will see the simulated stuffs...