rogeriobonatti / mushr_rhc

Receding Horizon Control

This module hosts the RHC controller first implemented on MuSHR stack. It is a model predictive contoller that plans to waypoints from a goal (instead of a reference trajectory). This controller is suitable for cars that don't have a planning module, but want simple MPC.

Tutorial

The following tutorial goes through installing/running the car.

Installing on the car

Note: if you are using the mushr image you can just clone the repo into ~/catkin_ws/src and it should work out of the box

Get pip:

sudo apt install python-pip

To run this module on the car, you need a few packages. To get them download the wheel file for torch from nvidia:

$ wget https://nvidia.box.com/shared/static/m6vy0c7rs8t1alrt9dqf7yt1z587d1jk.whl -O torch-1.1.0a0+b457266-cp27-cp27mu-linux_aarch64.whl
$ pip install torch-1.1.0a0+b457266-cp27-cp27mu-linux_aarch64.whl

Then get the future package:

pip install future

Then get the python packages necessary:

$ sudo apt install python-scipy
$ sudo apt install python-networkx
$ sudo apt install python-sklearn

librhc Layout

librhc (mushr_rhc_ros/src/librhc) is the core MPC code, with the other source being ROS interfacing code. The main components are:

• Cost function (librhc/cost): Takes into account the cost-to-go, collisions and other information to produce a cost for a set of trajectories.
• Model (librhc/model): A model of the car, currenly using the kinematic bicycle model.
• Trajectory generation (librhc/trajgen): Strategies for generating trajectory libraries for MPC to evaluate.
• Value function (librhc/value): Evaluation of positions of the car with resepct to a goal.
• World Representation (librhc/workrep): An occupancy grid based representation for the map.

mushr_rhc_ros ROS API

Publishers

Topic	Type	Description
/rhcontroller/markers	visualization_msgs/Marker	Halton points sampled in the map (for debugging purposes).
/rhcontroller/traj_chosen	geometry_msgs/PoseArray	The lowest cost trajectory (for debuggin purposes).
/car/mux/ackermann_cmd_mux/input/navigation	ackermann_msgs/AckermannDriveStamped	The lowest cost control to apply on the car.

Subscribers

Topic	Type	Description
/map_metadata	nav_msgs/MapMetaData	Uses dimension and resolution to create occupancy grid.
/move_base_simple/goal	geometry_msgs/PoseStamped	Goal to compute path to.
/car/car_pose	geometry_msgs/PoseStamped	When using simulated car pose Current pose of the car.
/car/pf/inferred_pose	geometry_msgs/PoseStamped	When using particle filter for localization Current pose of the car.

Services

Topic	Type	Description
/rhcontroller/reset/hard	std_srvs/Empty	Creates a new instance of the MPC object, redoing all initialization computation.
/rhcontroller/reset/soft	std_srvs/Empty	Resets parameters only, not redoing initialization.