【Apollo学习笔记】——规划模块TASK之PATH_REUSE_DECIDER

yuan〇2023-08-23 15:39

文章目录

前言

在Apollo星火计划学习笔记------Apollo路径规划算法原理与实践与【Apollo学习笔记】------Planning模块讲到......Stage::Process的PlanOnReferenceLine函数会依次调用task_list中的TASK，本文将会继续以LaneFollow为例依次介绍其中的TASK部分究竟做了哪些工作。由于个人能力所限，文章可能有纰漏的地方，还请批评斧正。

在modules/planning/conf/scenario/lane_follow_config.pb.txt配置文件中，我们可以看到LaneFollow所需要执行的所有task。

cpp 复制代码 
stage_config: {
  stage_type: LANE_FOLLOW_DEFAULT_STAGE
  enabled: true
  task_type: LANE_CHANGE_DECIDER
  task_type: PATH_REUSE_DECIDER
  task_type: PATH_LANE_BORROW_DECIDER
  task_type: PATH_BOUNDS_DECIDER
  task_type: PIECEWISE_JERK_PATH_OPTIMIZER
  task_type: PATH_ASSESSMENT_DECIDER
  task_type: PATH_DECIDER
  task_type: RULE_BASED_STOP_DECIDER
  task_type: SPEED_BOUNDS_PRIORI_DECIDER
  task_type: SPEED_HEURISTIC_OPTIMIZER
  task_type: SPEED_DECIDER
  task_type: SPEED_BOUNDS_FINAL_DECIDER
  task_type: PIECEWISE_JERK_SPEED_OPTIMIZER
  # task_type: PIECEWISE_JERK_NONLINEAR_SPEED_OPTIMIZER
  task_type: RSS_DECIDER

本文将继续介绍LaneFollow的第二个TASK------PATH_REUSE_DECIDER

PATH_REUSE_DECIDER功能简介


主要功能 ：检查路径是否可重用,提高帧间平顺性。
主要逻辑：主要判断是否可以重用上一帧规划的路径。若上一帧的路径未与障碍物发生碰撞，则可以重用，提高稳定性，节省计算量。若上一帧的规划出的路径发生碰撞，则重新规划路径。

PATH_REUSE_DECIDER相关配置

PATH_REUSE_DECIDER的相关配置集中在以下两个文件：modules/planning/conf/planning_config.pb.txt和modules/planning/conf/scenario/lane_follow_config.pb.txt

cpp 复制代码 
// modules/planning/conf/planning_config.pb.txt
default_task_config: {
  task_type: PATH_REUSE_DECIDER
  path_reuse_decider_config {
    reuse_path: false
  }
}
// modules/planning/conf/scenario/lane_follow_config.pb.txt
  task_config: {
    task_type: PATH_REUSE_DECIDER
    path_reuse_decider_config {
      reuse_path: false
    }
  }

可以看到，默认情况不启用PATH_REUSE，改为true后启用。

PATH_REUSE_DECIDER总体流程

接着来看一看PATH_REUSE_DECIDER的代码逻辑。代码路径：modules/planning/tasks/deciders/path_reuse_decider/path_reuse_decider.cc

主函数逻辑集中在Process函数中：

cpp 复制代码 
Status PathReuseDecider::Process(Frame* const frame,
                                 ReferenceLineInfo* const reference_line_info) {
  // Sanity checks.
  CHECK_NOTNULL(frame);
  CHECK_NOTNULL(reference_line_info);

  if (!Decider::config_.path_reuse_decider_config().reuse_path()) {
    ADEBUG << "skipping reusing path: conf";
    reference_line_info->set_path_reusable(false);
    return Status::OK();
  }

  // skip path reuse if not in LANE_FOLLOW_SCENARIO
  const auto scenario_type = injector_->planning_context()
                                 ->planning_status()
                                 .scenario()
                                 .scenario_type();
  if (scenario_type != ScenarioType::LANE_FOLLOW) {
    ADEBUG << "skipping reusing path: not in LANE_FOLLOW scenario";
    reference_line_info->set_path_reusable(false);
    return Status::OK();
  }

  // active path reuse during change_lane only
  auto* lane_change_status = injector_->planning_context()
                                 ->mutable_planning_status()
                                 ->mutable_change_lane();
  ADEBUG << "lane change status: " << lane_change_status->ShortDebugString();

  // skip path reuse if not in_change_lane
  if (lane_change_status->status() != ChangeLaneStatus::IN_CHANGE_LANE &&
      !FLAGS_enable_reuse_path_in_lane_follow) {
    ADEBUG << "skipping reusing path: not in lane_change";
    reference_line_info->set_path_reusable(false);
    return Status::OK();
  }

  // for hybrid model: skip reuse path for valid path reference
  const bool valid_model_output =
      reference_line_info->path_data().is_valid_path_reference();
  if (valid_model_output) {
    ADEBUG << "skipping reusing path: path reference is valid";
    reference_line_info->set_path_reusable(false);
    return Status::OK();
  }

  /*count total_path_ when in_change_lane && reuse_path*/
  ++total_path_counter_;

  /*reuse path when in non_change_lane reference line or
    optimization succeeded in change_lane reference line
  */
  bool is_change_lane_path = reference_line_info->IsChangeLanePath();
  if (is_change_lane_path && !lane_change_status->is_current_opt_succeed()) {
    reference_line_info->set_path_reusable(false);
    ADEBUG << "reusable_path_counter[" << reusable_path_counter_
           << "] total_path_counter[" << total_path_counter_ << "]";
    ADEBUG << "Stop reusing path when optimization failed on change lane path";
    return Status::OK();
  }

  // stop reusing current path:
  // 1. replan path
  // 2. collision
  // 3. failed to trim previous path
  // 4. speed optimization failed on previous path
  bool speed_optimization_successful = false;
  const auto& history_frame = injector_->frame_history()->Latest();
  if (history_frame) {
    const auto history_trajectory_type =
        history_frame->reference_line_info().front().trajectory_type();
    speed_optimization_successful =
        (history_trajectory_type != ADCTrajectory::SPEED_FALLBACK);
  }

  // const auto history_trajectory_type = injector_->FrameHistory()s
  //                                          ->Latest()
  //                                          ->reference_line_info()
  //                                          .front()
  //                                          .trajectory_type();
  if (path_reusable_) {
    if (!frame->current_frame_planned_trajectory().is_replan() &&
        speed_optimization_successful && IsCollisionFree(reference_line_info) &&
        TrimHistoryPath(frame, reference_line_info)) {
      ADEBUG << "reuse path";
      ++reusable_path_counter_;  // count reusable path
    } else {
      // stop reuse path
      ADEBUG << "stop reuse path";
      path_reusable_ = false;
    }
  } else {
    // F -> T
    auto* mutable_path_decider_status = injector_->planning_context()
                                            ->mutable_planning_status()
                                            ->mutable_path_decider();
    static constexpr int kWaitCycle = -2;  // wait 2 cycle

    const int front_static_obstacle_cycle_counter =
        mutable_path_decider_status->front_static_obstacle_cycle_counter();
    const bool ignore_blocking_obstacle =
        IsIgnoredBlockingObstacle(reference_line_info);
    ADEBUG << "counter[" << front_static_obstacle_cycle_counter
           << "] IsIgnoredBlockingObstacle[" << ignore_blocking_obstacle << "]";
    // stop reusing current path:
    // 1. blocking obstacle disappeared or moving far away
    // 2. trimming successful
    // 3. no statical obstacle collision.
    if ((front_static_obstacle_cycle_counter <= kWaitCycle ||
         ignore_blocking_obstacle) &&
        speed_optimization_successful && IsCollisionFree(reference_line_info) &&
        TrimHistoryPath(frame, reference_line_info)) {
      // enable reuse path
      ADEBUG << "reuse path: front_blocking_obstacle ignorable";
      path_reusable_ = true;
      ++reusable_path_counter_;
    }
  }

  reference_line_info->set_path_reusable(path_reusable_);
  ADEBUG << "reusable_path_counter[" << reusable_path_counter_
         << "] total_path_counter[" << total_path_counter_ << "]";
  return Status::OK();
}

PATH_REUSE_DECIDER相关子函数

IsCollisionFree

cpp 复制代码 
bool PathReuseDecider::IsCollisionFree(
    ReferenceLineInfo* const reference_line_info) {
  const ReferenceLine& reference_line = reference_line_info->reference_line();
  static constexpr double kMinObstacleArea = 1e-4;
  const double kSBuffer = 0.5;
  static constexpr int kNumExtraTailBoundPoint = 21;
  static constexpr double kPathBoundsDeciderResolution = 0.5;
  // current vehicle sl position
  common::SLPoint adc_position_sl;
  GetADCSLPoint(reference_line, &adc_position_sl);

  // current obstacles
  std::vector<Polygon2d> obstacle_polygons;
  for (auto obstacle :
       reference_line_info->path_decision()->obstacles().Items()) {
    // filtered all non-static objects and virtual obstacle
    if (!obstacle->IsStatic() || obstacle->IsVirtual()) {
      if (!obstacle->IsStatic()) {
        ADEBUG << "SPOT a dynamic obstacle";
      }
      if (obstacle->IsVirtual()) {
        ADEBUG << "SPOT a virtual obstacle";
      }
      continue;
    }

    const auto& obstacle_sl = obstacle->PerceptionSLBoundary();
    // Ignore obstacles behind ADC
    if ((obstacle_sl.end_s() < adc_position_sl.s() - kSBuffer) ||
        // Ignore too small obstacles.
        (obstacle_sl.end_s() - obstacle_sl.start_s()) *
                (obstacle_sl.end_l() - obstacle_sl.start_l()) <
            kMinObstacleArea) {
      continue;
    }
    obstacle_polygons.push_back(
        Polygon2d({Vec2d(obstacle_sl.start_s(), obstacle_sl.start_l()),
                   Vec2d(obstacle_sl.start_s(), obstacle_sl.end_l()),
                   Vec2d(obstacle_sl.end_s(), obstacle_sl.end_l()),
                   Vec2d(obstacle_sl.end_s(), obstacle_sl.start_l())}));
  }

  if (obstacle_polygons.empty()) {
    return true;
  }

  const auto& history_frame = injector_->frame_history()->Latest();
  if (!history_frame) {
    return false;
  }
  const DiscretizedPath& history_path =
      history_frame->current_frame_planned_path();
  // path end point
  // 将上一段轨迹的终点投影到SL坐标系下
  common::SLPoint path_end_position_sl;
  common::math::Vec2d path_end_position = {history_path.back().x(),
                                           history_path.back().y()};
  reference_line.XYToSL(path_end_position, &path_end_position_sl);
  for (size_t i = 0; i < history_path.size(); ++i) {
    common::SLPoint path_position_sl;
    common::math::Vec2d path_position = {history_path[i].x(),
                                         history_path[i].y()};
    reference_line.XYToSL(path_position, &path_position_sl);
    if (path_end_position_sl.s() - path_position_sl.s() <=
        kNumExtraTailBoundPoint * kPathBoundsDeciderResolution) {
      break;
    }
    if (path_position_sl.s() < adc_position_sl.s() - kSBuffer) {
      continue;
    }
    const auto& vehicle_box =
        common::VehicleConfigHelper::Instance()->GetBoundingBox(
            history_path[i]);
    std::vector<Vec2d> ABCDpoints = vehicle_box.GetAllCorners();
    for (const auto& corner_point : ABCDpoints) {
      // For each corner point, project it onto reference_line
      common::SLPoint curr_point_sl;
      if (!reference_line.XYToSL(corner_point, &curr_point_sl)) {
        AERROR << "Failed to get the projection from point onto "
                  "reference_line";
        return false;
      }
      auto curr_point = Vec2d(curr_point_sl.s(), curr_point_sl.l());
      // Check if it's in any polygon of other static obstacles.
      for (const auto& obstacle_polygon : obstacle_polygons) {
        if (obstacle_polygon.IsPointIn(curr_point)) {
          // for debug
          ADEBUG << "s distance to end point:" << path_end_position_sl.s();
          ADEBUG << "s distance to end point:" << path_position_sl.s();
          ADEBUG << "[" << i << "]"
                 << ", history_path[i].x(): " << std::setprecision(9)
                 << history_path[i].x() << ", history_path[i].y()"
                 << std::setprecision(9) << history_path[i].y();
          ADEBUG << "collision:" << curr_point.x() << ", " << curr_point.y();
          Vec2d xy_point;
          reference_line.SLToXY(curr_point_sl, &xy_point);
          ADEBUG << "collision:" << xy_point.x() << ", " << xy_point.y();

          return false;
        }
      }
    }
  }
  return true;
}

TrimHistoryPath

cpp 复制代码 
bool PathReuseDecider::TrimHistoryPath(
    Frame* frame, ReferenceLineInfo* const reference_line_info) {
  const ReferenceLine& reference_line = reference_line_info->reference_line();
  const auto& history_frame = injector_->frame_history()->Latest();
  if (!history_frame) {
    ADEBUG << "no history frame";
    return false;
  }
  // 找到上一帧轨迹的起始点
  const common::TrajectoryPoint history_planning_start_point =
      history_frame->PlanningStartPoint();
  common::PathPoint history_init_path_point =
      history_planning_start_point.path_point();
  ADEBUG << "history_init_path_point x:[" << std::setprecision(9)
         << history_init_path_point.x() << "], y["
         << history_init_path_point.y() << "], s: ["
         << history_init_path_point.s() << "]";
  // 当前周期规划的起点
  const common::TrajectoryPoint planning_start_point =
      frame->PlanningStartPoint();
  common::PathPoint init_path_point = planning_start_point.path_point();
  ADEBUG << "init_path_point x:[" << std::setprecision(9) << init_path_point.x()
         << "], y[" << init_path_point.y() << "], s: [" << init_path_point.s()
         << "]";
  
  const DiscretizedPath& history_path =
      history_frame->current_frame_planned_path();
  DiscretizedPath trimmed_path;
  // 获取自车的SL坐标
  common::SLPoint adc_position_sl;  // current vehicle sl position
  GetADCSLPoint(reference_line, &adc_position_sl);
  ADEBUG << "adc_position_sl.s(): " << adc_position_sl.s();

  size_t path_start_index = 0;

  for (size_t i = 0; i < history_path.size(); ++i) {
    // find previous init point
    // 找到上周期轨迹规划的起点索引
    if (history_path[i].s() > 0) {
      path_start_index = i;
      break;
    }
  }
  ADEBUG << "!!!path_start_index[" << path_start_index << "]";

  // get current s=0
  common::SLPoint init_path_position_sl;
  // 当前轨迹的起点
  reference_line.XYToSL(init_path_point, &init_path_position_sl);
  bool inserted_init_point = false;
  //匹配当前规划起点位置，裁剪该点之后的轨迹
  for (size_t i = path_start_index; i < history_path.size(); ++i) {
    common::SLPoint path_position_sl;
    common::math::Vec2d path_position = {history_path[i].x(),
                                         history_path[i].y()};

    reference_line.XYToSL(path_position, &path_position_sl);

    double updated_s = path_position_sl.s() - init_path_position_sl.s();
    // insert init point
    if (updated_s > 0 && !inserted_init_point) {
      trimmed_path.emplace_back(init_path_point);
      trimmed_path.back().set_s(0);
      inserted_init_point = true;
    }

    trimmed_path.emplace_back(history_path[i]);

    // if (i < 50) {
    //   ADEBUG << "path_point:[" << i << "]" << updated_s;
    //   path_position_sl.s();
    //   ADEBUG << std::setprecision(9) << "path_point:[" << i << "]"
    //          << "x: [" << history_path[i].x() << "], y:[" <<
    //          history_path[i].y()
    //          << "]. s[" << history_path[i].s() << "]";
    // }
    trimmed_path.back().set_s(updated_s);
  }

  ADEBUG << "trimmed_path[0]: " << trimmed_path.front().s();
  ADEBUG << "[END] trimmed_path.size(): " << trimmed_path.size();
  // 检查裁剪出来的轨迹是不是过短
  if (!NotShortPath(trimmed_path)) {
    ADEBUG << "short path: " << trimmed_path.size();
    return false;
  }

  // set path
  auto path_data = reference_line_info->mutable_path_data();
  ADEBUG << "previous path_data size: " << history_path.size();
  path_data->SetReferenceLine(&reference_line);
  ADEBUG << "previous path_data size: " << path_data->discretized_path().size();
  path_data->SetDiscretizedPath(DiscretizedPath(std::move(trimmed_path)));
  ADEBUG << "not short path: " << trimmed_path.size();
  ADEBUG << "current path size: "
         << reference_line_info->path_data().discretized_path().size();

  return true;
}

IsIgnoredBlockingObstacle和GetBlockingObstacleS

前方堵塞的障碍物是否离开足够远的距离

cpp 复制代码 
bool PathReuseDecider::IsIgnoredBlockingObstacle(
    ReferenceLineInfo* const reference_line_info) {
  const ReferenceLine& reference_line = reference_line_info->reference_line();
  static constexpr double kSDistBuffer = 30.0;  // meter
  static constexpr int kTimeBuffer = 3;         // second
  // vehicle speed
  double adc_speed = injector_->vehicle_state()->linear_velocity();
  double final_s_buffer = std::max(kSDistBuffer, kTimeBuffer * adc_speed);
  // current vehicle s position
  common::SLPoint adc_position_sl;
  GetADCSLPoint(reference_line, &adc_position_sl);
  // blocking obstacle start s
  double blocking_obstacle_start_s;
  if (GetBlockingObstacleS(reference_line_info, &blocking_obstacle_start_s) &&
      // distance to blocking obstacle
      (blocking_obstacle_start_s - adc_position_sl.s() > final_s_buffer)) {
    ADEBUG << "blocking obstacle distance: "
           << blocking_obstacle_start_s - adc_position_sl.s();
    return true;
  } else {
    return false;
  }
}
cpp 复制代码 
bool PathReuseDecider::GetBlockingObstacleS(
    ReferenceLineInfo* const reference_line_info, double* blocking_obstacle_s) {
  auto* mutable_path_decider_status = injector_->planning_context()
                                          ->mutable_planning_status()
                                          ->mutable_path_decider();
  // get blocking obstacle ID (front_static_obstacle_id)
  const std::string& blocking_obstacle_ID =
      mutable_path_decider_status->front_static_obstacle_id();
  const IndexedList<std::string, Obstacle>& indexed_obstacles =
      reference_line_info->path_decision()->obstacles();
  const auto* blocking_obstacle = indexed_obstacles.Find(blocking_obstacle_ID);

  if (blocking_obstacle == nullptr) {
    return false;
  }

  const auto& obstacle_sl = blocking_obstacle->PerceptionSLBoundary();
  *blocking_obstacle_s = obstacle_sl.start_s();
  ADEBUG << "blocking obstacle distance: " << obstacle_sl.start_s();
  return true;
}

Else

在启用reuse之后，之后的task会有这样一段代码，用以跳过以下流程，沿用之前的path

cpp 复制代码 
  // skip path_lane_borrow_decider if reused path
  if (FLAGS_enable_skip_path_tasks && reference_line_info->path_reusable()) {
    // for debug
    AINFO << "skip due to reusing path";
    return Status::OK();
  }

参考

[1] Apollo Planning决策规划代码详细解析 (7): PathReuseDecider

[2] Apollo6.0 PathReuseDecider流程与代码解析

相关推荐
小芒果_011 分钟前
P11229 [CSP-J 2024] 小木棍
c++·算法·信息学奥赛
qq_434085903 分钟前
Day 52 || 739. 每日温度 、 496.下一个更大元素 I 、503.下一个更大元素II
算法
Beau_Will3 分钟前
ZISUOJ 2024算法基础公选课练习一（2）
算法
XuanRanDev6 分钟前
【每日一题】LeetCode - 三数之和
数据结构·算法·leetcode·1024程序员节
gkdpjj7 分钟前
C++优选算法十 哈希表
c++·算法·散列表
代码猪猪傻瓜coding7 分钟前
力扣1 两数之和
数据结构·算法·leetcode
小码农<^_^>1 小时前
优选算法精品课--滑动窗口算法（一）
算法
羊小猪~~1 小时前
神经网络基础--什么是正向传播？？什么是方向传播？？
人工智能·pytorch·python·深度学习·神经网络·算法·机器学习
软工菜鸡1 小时前
预训练语言模型BERT——PaddleNLP中的预训练模型
大数据·人工智能·深度学习·算法·语言模型·自然语言处理·bert
南宫生1 小时前
贪心算法习题其三【力扣】【算法学习day.20】
java·数据结构·学习·算法·leetcode·贪心算法
热门推荐
01【HarmonyOS】HUAWEI DevEco Studio 下载地址汇总02(欧拉)openEuler系统添加网卡文件配置流程、(欧拉)openEuler系统手动配置ipv6地址流程、(欧拉)openEuler系统网络管理说明03组基轨迹建模 GBTM的介绍与实现（Stata 或 R）04【AIGC】重塑未来的科技巨轮05全面解析：构建基于深度学习的安全帽检测系统（UI界面+YOLO代码+数据集）06【经验分享】Ubuntu22.04安装微信(linux官方版)07基于YOLOv10深度学习的CT扫描图像肾结石智能检测系统【python源码+Pyqt5界面+数据集+训练代码】深度学习实战、目标检测08Ubuntu 20.04使用Livox mid 360 测试 FAST_LIO09RAG 实践- Ollama+RagFlow 部署本地知识库10【TC3xx芯片】TC3xx芯片电源管理系统PMS详解