Simulation Platform for Single/Multi Agent System

This is a repo for multi agent system simulation built upon MUJOCO. It now has several features:

1. Single/multi robots
   • Differential-drive Mobile Robot
   • A Scene Generator for generating any number of robot and particles
2. Sensing
   • Top Down Camera Calibration and Detection
   • Real-time Contour Reconstruction
3. Control
   • Keyboard Controller
   • PD Controller
   • MPC Controller
4. Planning
   • A* on 2D GridMap
   • RRTConnect(ompl)
   • kinodynamic RRT(ompl)
5. Learning
   • Reinforcement Learning

Installation (Ubuntu 22.04):

1. create your workspace

mkdir ws && cd ws

2. clone

git clone git@github.com:KANZEZ/m0.git

3. create your own virtual env

python3 -m venv your_env

4. in your virtual env, install the packages

pip3 install -e . 

5. run some examples in m0/examples folder to see if success

6. if you want to use mpc, please follow the install instruction for acados.

Usage & Demo

Differential-drive Mobile Robot

Differential-drive mobile robot.
State: [x, y, $\theta$]
Control input: [v, w]
Dynamics: [$\dot x$, $\dot y$, $\dot \theta$] = [v $cos\theta$, v $sin\theta$, w]

Sensing

A top-down camera is fixed above the playground and can detect the contour of the swarm of particles, and then we use FFT to parametrize and reconstruct the contour.

python3 examples/test_contour_recons.py

Control

1. PD control:

python3 examples/test_pid.py

2. MPC:

python3 examples/test_mpc.py

Planning

1. Multi-agent mppi

2. Geometry Planning(RRTConnect & A* on grid map)

python3 examples/test_astar.py

python3 examples/test_rrtc.py

3. Kinodynamic RRT(consider robot dynamics)

python3 examples/test_kinorrt.py

Learning

1. RL

python3 examples/test_rl.py