ACRE Control Go2 Package

Introduction

The acre_ctrl package is a ROS2 package used to write and run control algorithms for the Unitree Go2. Before using acre_ctrl, please familiarize yourself with the basics of using the Unitree Go2.

Note

Only the /cmd_vel and /sport_odom topics were included in this diagram. All sensor subscribers connect to corresponding topics. These topics can be edited in the node configuration.

Using acre_ctrl

Sportmode Driver

All control algorithms communicate with the sportmode_driver node. This node is responsible for communicating with the Unitree SDK and executing control commands on the Go2. To start the sportmode_driver node run:

ros2 run acre_ctrl sportmode_driver

sportmode_driver params:

Parameter	Default	Options	Description
gait	classic	Classic or Agile	The Go2's gait. See the Go2 Docs for more information

Using Python Algorithms with the Control Node

acre_ctrl allows users to write simple algorithm classes in Python and then dynamically load these classes into the control loop to be executed. An algorithm class can be executed using the following command. Note that generally you should run the sportmode_driver and then the control_node.

ros2 run acre_ctrl control_node --ros-args -p algorithm:=/path/to/algorithm.py

control_node params:

Parameter	Default	Description
algorithm	None	The path to the control algorithm to execute
frequency(Hz)	10.0	The frequency to run the control algorithm at
pose_topic	/mocap_pose	Pose topic name
odom_topic	/sport_odom	Odometry topic name
map_topic	/map	Map topic name
point_cloud_topic	/unitree/slam_lidar/points	Point Cloud topic name
imu_topic	/unitree/slam_lidar/imu	IMU topic name
path_topic	/plan	Plan topic name
traj_topic	/joint_trajectory	Trajectory topic name

Spin Example

This example uses the acre_ctrl stack to make the Go2 spin in place.

ros2 run acre_ctrl sportmode_driver --ros-args -p gait:="classic"
ros2 run acre_ctrl control_node --ros-args -p algorithm:=/workspace/src/acre_ctrl/example/spin_example.py

Writing Algorithms for acre_ctrl

Python Algorithms

acre_ctrl provides a convient interface for writing control algorithms in python. To write a python control algorithm, you must create a class that inherits from the ControlAlgorithm class.

Control algorithms must implement a compute function

class ControlAlgorithm(ABC):
    def init(self) -> None:
        pass

    @abstractmethod
    def compute(self, input: ComponentRegistry) -> Twist:
        ...

The input to the compute function is a ComponentRegistry. This class stores all the possible inputs available for algorithm authors. When writing an algorithm you can access the input components you need using ROS2 types.

@dataclass
class ComponentRegistry:
    dt:             Optional[float]             = None
    pose:           Optional[Pose]              = None
    odom:           Optional[Odometry]          = None
    goal:           Optional[Pose]              = None
    path:           Optional[Path]              = None
    traj:           Optional[JointTrajectory]   = None
    point_cloud:    Optional[PointCloud2]       = None
    map:            Optional[OccupancyGrid]     = None
    imu:            Optional[Imu]               = None

To define which of the inputs you algorithm uses, you must use the @components(...) class decorator. This decorator tells acre_ctrl which inputs your algorithm expects. The naming conventions for the decorator follow the names of the ComponentRegistry fields.

@components("odom", "goal", "map")

spin_example.py algorithm

To understand how to write algorithm code for acre_ctrl refer to this simple example:

from acre_ctrl.algorithm import ComponentRegistry, ControlAlgorithm, components
from geometry_msgs.msg import Twist

@components("odom")
class MyAlgo(ControlAlgorithm):
    
    def compute(self, input: ComponentRegistry) -> Twist:
        print(f"Current linear_x{input.odom.twist.twist.linear.x}, linear_y{input.odom.twist.twist.linear.y}")
        twist = Twist()
        twist.angular.z = 0.5

        return twist

C++ Algorithms

Unlike Python algorithms, C++ algorithms cannot be loaded dynamically using the control_node. If you wish to write a C++ algorithm you should write a C++ ROS2 node that communicates directly with the sportmode_driver.