Ubuntu20.04，ROS Neotic。Move It！（8）运动规划API Motion Planning API
作者：互联网
在MoveIt中，使用插件基础结构来加载运动计划器。这样，MoveIt可以在运行时加载运动计划器。在此示例中，我们将遍历执行此操作所需的C ++代码。
运行演示
打开两个终端。在第一个终端程序中，启动RViz并等待所有内容完成加载：
roslaunch panda_moveit_config demo.launch
在第二个终端中，运行启动文件：
roslaunch moveit_tutorials motion_planning_api_tutorial.launch
在RViz中，我们应该能够看到最终重播了四个轨迹：
机器人将手臂移动到第一个目标姿态，
机器人将手臂移至关节目标，
机器人将手臂移回原始目标姿态，
机器人将手臂移动到新的目标姿态，同时保持末端执行器水平。
开始
设置以开始使用计划程序非常简单。计划程序在MoveIt中设置为插件，您可以使用ROS pluginlib接口来加载要使用的任何计划程序。在加载计划器之前，我们需要两个对象，即RobotModel和PlanningScene。我们将从实例化RobotModelLoader 对象开始，该对象将在ROS参数服务器上查找机器人说明，并构造供我们使用的RobotModel。
const std::string PLANNING_GROUP = "panda_arm";
robot_model_loader::RobotModelLoader robot_model_loader("robot_description");
const moveit::core::RobotModelPtr& robot_model = robot_model_loader.getModel();
/* Create a RobotState and JointModelGroup to keep track of the current robot pose and planning group*/
moveit::core::RobotStatePtr robot_state(new moveit::core::RobotState(robot_model));
const moveit::core::JointModelGroup* joint_model_group = robot_state->getJointModelGroup(PLANNING_GROUP);
使用RobotModel，我们可以构造一个 可维护世界（包括机器人）状态的PlanningScene。
planning_scene::PlanningScenePtr planning_scene(new planning_scene::PlanningScene(robot_model));
配置有效的机械手状态
planning_scene->getCurrentStateNonConst().setToDefaultValues(joint_model_group, "ready");
现在，我们将构造一个加载器以按名称加载规划器。请注意，我们在这里使用ROS pluginlib库。
boost::scoped_ptr<pluginlib::ClassLoader<planning_interface::PlannerManager>> planner_plugin_loader;
planning_interface::PlannerManagerPtr planner_instance;
std::string planner_plugin_name;
我们将从ROS参数服务器中获取要加载的计划插件的名称，然后加载计划程序，确保捕获所有异常。
if (!node_handle.getParam("planning_plugin", planner_plugin_name))
ROS_FATAL_STREAM("Could not find planner plugin name");
try
{
planner_plugin_loader.reset(new pluginlib::ClassLoader<planning_interface::PlannerManager>(
"moveit_core", "planning_interface::PlannerManager"));
}
catch (pluginlib::PluginlibException& ex)
{
ROS_FATAL_STREAM("Exception while creating planning plugin loader " << ex.what());
}
try
{
planner_instance.reset(planner_plugin_loader->createUnmanagedInstance(planner_plugin_name));
if (!planner_instance->initialize(robot_model, node_handle.getNamespace()))
ROS_FATAL_STREAM("Could not initialize planner instance");
ROS_INFO_STREAM("Using planning interface '" << planner_instance->getDescription() << "'");
}
catch (pluginlib::PluginlibException& ex)
{
const std::vector<std::string>& classes = planner_plugin_loader->getDeclaredClasses();
std::stringstream ss;
for (const auto& cls : classes)
ss << cls << " ";
ROS_ERROR_STREAM("Exception while loading planner '" << planner_plugin_name << "': " << ex.what() << std::endl
<< "Available plugins: " << ss.str());
}
可视化
软件包MoveItVisualTools提供了许多功能，用于可视化RViz中的对象，机械手和轨迹以及调试工具，例如脚本的逐步自省。
namespace rvt = rviz_visual_tools;
moveit_visual_tools::MoveItVisualTools visual_tools("panda_link0");
visual_tools.loadRobotStatePub("/display_robot_state");
visual_tools.enableBatchPublishing();
visual_tools.deleteAllMarkers(); // clear all old markers
visual_tools.trigger();
/* Remote control is an introspection tool that allows users to step through a high level script
via buttons and keyboard shortcuts in RViz */
visual_tools.loadRemoteControl();
/* RViz provides many types of markers, in this demo we will use text, cylinders, and spheres*/
Eigen::Isometry3d text_pose = Eigen::Isometry3d::Identity();
text_pose.translation().z() = 1.75;
visual_tools.publishText(text_pose, "Motion Planning API Demo", rvt::WHITE, rvt::XLARGE);
/* Batch publishing is used to reduce the number of messages being sent to RViz for large visualizations */
visual_tools.trigger();
/* We can also use visual_tools to wait for user input */
visual_tools.prompt("Press 'next' in the RvizVisualToolsGui window to start the demo");
目标姿势
现在，我们将为熊猫的手臂创建一个运动计划请求，并指定所需的末端执行器姿势作为输入。
visual_tools.publishRobotState(planning_scene->getCurrentStateNonConst(), rviz_visual_tools::GREEN);
visual_tools.trigger();
planning_interface::MotionPlanRequest req;
planning_interface::MotionPlanResponse res;
geometry_msgs::PoseStamped pose;
pose.header.frame_id = "panda_link0";
pose.pose.position.x = 0.3;
pose.pose.position.y = 0.4;
pose.pose.position.z = 0.75;
pose.pose.orientation.w = 1.0;
位置公差为0.01 m，方向公差为0.01弧度
std::vector<double> tolerance_pose(3, 0.01);
std::vector<double> tolerance_angle(3, 0.01);
我们将使用kinematic_constraints 包中提供的帮助器函数将请求创建为约束 。
moveit_msgs::Constraints pose_goal =
kinematic_constraints::constructGoalConstraints("panda_link8", pose, tolerance_pose, tolerance_angle);
req.group_name = PLANNING_GROUP;
req.goal_constraints.push_back(pose_goal);
现在，我们构建一个计划context ，该context 封装了场景，请求和响应。我们使用此计划context 致电计划者
planning_interface::PlanningContextPtr context =
planner_instance->getPlanningContext(planning_scene, req, res.error_code_);
context->solve(res);
if (res.error_code_.val != res.error_code_.SUCCESS)
{
ROS_ERROR("Could not compute plan successfully");
return 0;
}
可视化结果
ros::Publisher display_publisher =
node_handle.advertise<moveit_msgs::DisplayTrajectory>("/move_group/display_planned_path", 1, true);
moveit_msgs::DisplayTrajectory display_trajectory;
/* Visualize the trajectory */
moveit_msgs::MotionPlanResponse response;
res.getMessage(response);
display_trajectory.trajectory_start = response.trajectory_start;
display_trajectory.trajectory.push_back(response.trajectory);
visual_tools.publishTrajectoryLine(display_trajectory.trajectory.back(), joint_model_group);
visual_tools.trigger();
display_publisher.publish(display_trajectory);
/* Set the state in the planning scene to the final state of the last plan */
robot_state->setJointGroupPositions(joint_model_group, response.trajectory.joint_trajectory.points.back().positions);
planning_scene->setCurrentState(*robot_state.get());
显示目标状态
visual_tools.publishRobotState(planning_scene->getCurrentStateNonConst(), rviz_visual_tools::GREEN);
visual_tools.publishAxisLabeled(pose.pose, "goal_1");
visual_tools.publishText(text_pose, "Pose Goal (1)", rvt::WHITE, rvt::XLARGE);
visual_tools.trigger();
/* We can also use visual_tools to wait for user input */
visual_tools.prompt("Press 'next' in the RvizVisualToolsGui window to continue the demo");
关节空间目标
现在，设定一个联合太空目标
moveit::core::RobotState goal_state(robot_model);
std::vector<double> joint_values = { -1.0, 0.7, 0.7, -1.5, -0.7, 2.0, 0.0 };
goal_state.setJointGroupPositions(joint_model_group, joint_values);
moveit_msgs::Constraints joint_goal = kinematic_constraints::constructGoalConstraints(goal_state, joint_model_group);
req.goal_constraints.clear();
req.goal_constraints.push_back(joint_goal);
call规划器planer并可视化轨迹
/* Re-construct the planning context */
context = planner_instance->getPlanningContext(planning_scene, req, res.error_code_);
/* Call the Planner */
context->solve(res);
/* Check that the planning was successful */
if (res.error_code_.val != res.error_code_.SUCCESS)
{
ROS_ERROR("Could not compute plan successfully");
return 0;
}
/* Visualize the trajectory */
res.getMessage(response);
display_trajectory.trajectory.push_back(response.trajectory);
/* Now you should see two planned trajectories in series*/
visual_tools.publishTrajectoryLine(display_trajectory.trajectory.back(), joint_model_group);
visual_tools.trigger();
display_publisher.publish(display_trajectory);
/* We will add more goals. But first, set the state in the planning
scene to the final state of the last plan */
robot_state->setJointGroupPositions(joint_model_group, response.trajectory.joint_trajectory.points.back().positions);
planning_scene->setCurrentState(*robot_state.get());
显示目标状态
visual_tools.publishRobotState(planning_scene->getCurrentStateNonConst(), rviz_visual_tools::GREEN);
visual_tools.publishAxisLabeled(pose.pose, "goal_2");
visual_tools.publishText(text_pose, "Joint Space Goal (2)", rvt::WHITE, rvt::XLARGE);
visual_tools.trigger();
/* Wait for user input */
visual_tools.prompt("Press 'next' in the RvizVisualToolsGui window to continue the demo");
/* Now, we go back to the first goal to prepare for orientation constrained planning */
req.goal_constraints.clear();
req.goal_constraints.push_back(pose_goal);
context = planner_instance->getPlanningContext(planning_scene, req, res.error_code_);
context->solve(res);
res.getMessage(response);
display_trajectory.trajectory.push_back(response.trajectory);
visual_tools.publishTrajectoryLine(display_trajectory.trajectory.back(), joint_model_group);
visual_tools.trigger();
display_publisher.publish(display_trajectory);
/* Set the state in the planning scene to the final state of the last plan */
robot_state->setJointGroupPositions(joint_model_group, response.trajectory.joint_trajectory.points.back().positions);
planning_scene->setCurrentState(*robot_state.get());
显示目标状态
visual_tools.publishRobotState(planning_scene->getCurrentStateNonConst(), rviz_visual_tools::GREEN);
visual_tools.trigger();
/* Wait for user input */
visual_tools.prompt("Press 'next' in the RvizVisualToolsGui window to continue the demo");
添加路径约束
让我们再次添加一个新的姿势目标。这次我们还将向运动添加路径约束。
/* Let's create a new pose goal */
pose.pose.position.x = 0.32;
pose.pose.position.y = -0.25;
pose.pose.position.z = 0.65;
pose.pose.orientation.w = 1.0;
moveit_msgs::Constraints pose_goal_2 =
kinematic_constraints::constructGoalConstraints("panda_link8", pose, tolerance_pose, tolerance_angle);
/* Now, let's try to move to this new pose goal*/
req.goal_constraints.clear();
req.goal_constraints.push_back(pose_goal_2);
/* But, let's impose a path constraint on the motion.
Here, we are asking for the end-effector to stay level*/
geometry_msgs::QuaternionStamped quaternion;
quaternion.header.frame_id = "panda_link0";
quaternion.quaternion.w = 1.0;
req.path_constraints = kinematic_constraints::constructGoalConstraints("panda_link8", quaternion);
由于施加了路径约束，规划者需要在末端执行器（机器人的工作空间）的可能位置空间中进行推理，因此，我们还需要为允许的规划量指定一个界限。注意：默认界限由WorkspaceBounds请求适配器自动填充（属于OMPL管道的一部分，但在本示例中未使用）。我们使用的界限肯定包括手臂的可到达空间。这很好，因为在规划手臂时不会在此卷中进行采样；边界仅用于确定采样的配置是否有效。
req.workspace_parameters.min_corner.x = req.workspace_parameters.min_corner.y =
req.workspace_parameters.min_corner.z = -2.0;
req.workspace_parameters.max_corner.x = req.workspace_parameters.max_corner.y =
req.workspace_parameters.max_corner.z = 2.0;
call planer规划器，并可视化到目前为止创建的所有计划。
context = planner_instance->getPlanningContext(planning_scene, req, res.error_code_);
context->solve(res);
res.getMessage(response);
display_trajectory.trajectory.push_back(response.trajectory);
visual_tools.publishTrajectoryLine(display_trajectory.trajectory.back(), joint_model_group);
visual_tools.trigger();
display_publisher.publish(display_trajectory);
/* Set the state in the planning scene to the final state of the last plan */
robot_state->setJointGroupPositions(joint_model_group, response.trajectory.joint_trajectory.points.back().positions);
planning_scene->setCurrentState(*robot_state.get());
显示目标状态
visual_tools.publishRobotState(planning_scene->getCurrentStateNonConst(), rviz_visual_tools::GREEN);
visual_tools.publishAxisLabeled(pose.pose, "goal_3");
visual_tools.publishText(text_pose, "Orientation Constrained Motion Plan (3)", rvt::WHITE, rvt::XLARGE);
visual_tools.trigger();
标签：Ubuntu20.04,goal,trajectory,planning,pose,Planning,visual,API,tools 来源： https://blog.csdn.net/qq_33328642/article/details/116926375