README.md

slamrs

This repository contains the code for my attempt to overcome the issues faced with the old gridmap-slam-robot project. The issues were mainly related to performance of the algorithms and the absense of a possibility to publish for the web. I have since also upgraded the system with a LIDAR unit from a Neato robot vacuum cleaner, such as those described in ssloy/neato-xv11-lidar, which also increases the amount of data being available to be processed.

Current Features

• Home-grown immediate mode OpenGL shape drawing library using vertex buffers and shaders.

  • Rewrite and improved version of the renderer developed in antbern/gridmap-slam-robot, originally inspired by the ShapeRenderer in libgdx.

• A fully typed topic-based publish-subscribe system.

• A system of Nodes that communicate through the pub-sub system.

  • Fully declarative configuration file for enabling and connecting nodes to topics as well as configuring their parameters.
  • A simulator for a differential drive robot with a laser range scanner (a.k.a a LIDAR).
  • Ability to connect to and control the robot with a Neato LIDAR via the serial port or over a TCP connection.
  • Implementation of point-to-plane ICP for doing scan matching (link to videos and resources).
  • Implementation of grid-based SLAM using a particle filter and Bayesian log-odds update rules.
  • Fully flexible and customizable data visualization node.

• Runs on desktop and in the browser through wasm! (except for nodes that load files or connects to the serial port)

• Robot with a Neato XV11 LIDAR

  • Rewrite of gridmap-slam-robot project for the iLabs Challenger RP2040 WiFi/BLE MkII board containing a rp2040 dual-core processor and a esp32 companion chip for network capabilities.
  • Firmware written using rtic.

Future Directions

An unfiltered (and unstructured) list of high-level ideas for future development:

• A Node that can record and replay data on topics, with support for persistence and time-stamped messages.
• Improve configuration for sensor and robot motion models, bot forward and inverse.
• Improved pointmap map that can be used for more "global" mapping with ICP.
  • Need to drop points in a smart way to keep complexity of the ICP algorightm down.
  • Can we use this with a particle filter?
• Extraction of "landmarks" from lidar data
  • Perhaps including the strength values from the LIDAR unit, would need to be added to the simulator.
• Implement EKF or graph-based slam using the extracted landmarks.
• Add "exploration mode" where driving commands are generated to explore the world and gain more knowledge about it.
• Path generation and following by mouse click on the map.
• Add dynamic "Obstacles" to the Simulator.
• Saving and loading of maps (+global localization? could use particle filter with uniform starting locations for that)
• Connection with "real robot", updated from my previous gridmap-slam-robot project. (#99)
• Arbitrary input configuration for Nodes, i.e., node inputs can be connected to topics or constants. Same with outputs.
• Async? Can we make us of that in a good way?
  • Right now only reasonable when waiting for a value to arrive through the pub-sub system, but might be relevant still.