venkisagunner93 / beertender_robot

A repository to learn navigation in ROS and C++

beertender_robot

This repository is for building high-level software for beertender robot.

TO-DO:

• Create-2 driver integration
• Path planning
  • Global planning
  • Local planning
• Localization
  • Setup basic cartographer
• CI/CD
  • Build pipeline
  • Self-hosted runner
  • Testing pipeline
    • gtest/gmock
    • rostest
  • Code quality pipeline
    • clang-format
    • catkin_lint
    • roscpp_lint
  • Deployment pipeline
    • Containerize deployment
    • Robot as a docker registry
  • Data collection pipeline
    • Logs
    • Bag files
    • Metrics

How to run?

cd $HOME
git clone https://github.com/venkisagunner93/beertender_robot.git
cd beertender_robot
source tools/setup.sh
beertender --sdk
sdk

An rviz window should open up. Enjoy playing with create2 robot

ROS Graph

beertender_robot's Issues

Setup simulator using gazebo

Make a simulator:

• Setup gazebo model for create 2
• Setup a test world (my apartment)
• Deprecate existing simulation codebase
• Add lidar to create 2 gazebo model
• Clean up TF tree

Teleop capability

Add teleop capability such that the local planner should be preempted once the teleop kicks in.

Setup testing stage in CI pipeline

Current CI pipeline does not have any tests. Add a stage to run tests on the code base.

Simple structure for planner

Create a simple structure for planner to do the following:

• Able to get global goal
• Able to generate a global path from global goal provided
• Able to execute global path by providing it to local planner

Add a docker compose file to launch and tear down SW in robot

So far, the tools support pushing a docker image to the robot and pulling it locally in the robot's motherboard. But, the robot does not start its software automatically when a new updated image is pushed. In order for this to happen, the robot needs a docker_compose.yaml file that can architect containers spawning and tear down.

Improve laser simulation code

Laser simulation code is taking too much computational power right now. Modify code based on ray tracing approach. The idea is to use the co-ordinate transform between base_link and map frames and with a known occupancy grid map, we can simulate laser points. This repository implements ray tracer. Integrate the ray tracer with robot_sim class.

Unit tests for both global and local planners

Add unit tests for both local and global planner packages

Simple laser scan feature

Add a simple laser scan feature in sim pipeline

Document the steps to setup robot's operating system from scratch

There are few steps that needs to be followed to setup the robot's operating system:

• Create a basic OS for Nano
• Setup so that the robot has GUI turned off
• Install docker
• Setup a insecure docker registry

Create separate launch files for visualization in remote machine

Visualization tools shall go in a different launch file and if possible dockerize it with x11 forwarding.

Add motion compensation to lidar simulation

At the moment, the laser simulation is not robust to motion compensation. Here is a paper that shall be implemented to achieve good simulation performance: https://arxiv.org/pdf/1706.05886.pdf

Add linting to CI pipeline

Add roscpp_lint stage in the CI pipeline.

Deploy code as docker image

Binaries are being deployed as such. Needs to be deployed as docker images, where the images can be started from another machine remotely.

Refactor global_planner to have action server

Create global planner with action server.

##Current implementation:

• Subscribed to nav/global_goal topic and publishes global path to nav/global_path
• Not ideal way to do this. Should have preemption, abort, success and feedback capabilities

Add working version of cartographer to the SDK

Add cartographer to the SDK and create a configuration so that it works for a test environment.

Docker registry in robot

Reflash the operating system in robot's computer and install only docker. Modify the computer as a docker registry

Modify current URDF model to create2 urdf

The current URDF model is not representative of irobot create2's model. Change the URDF to represent irobot create2's model.

Complete simulation pipeline

Simulation is not implemented at the moment. Needs to be scoped out for future and implemented (at least barebones)

Implement pose graph from scratch

Notes:

• Create two new classes for vertices and edges
• Subscribe to odometry for a configurable sampling size and calculate relative motion composition
• Subscribe to laser scans and add it into the corresponding vertex after accumulating enough motion through odometry
• Perform local slam using ceres between successive vertices to keep initial cost of global slam low
• Perform global slam by detecting loop closure between two vertices where the euclidean distance is lower than a threshold
• Based on the diagonal elements of the reduced hessian matrix, determine which edges need to be adjusted to optimize totness of the edges so that the final answer is optimal

Add odom frame from wheel odometry

Add a odom frame in robot sim from wheel odometry for the robot. This should conform to REP 105 tf tree structure.