From 281345c3d9b4fc6261f03164cab5b8b14531077f Mon Sep 17 00:00:00 2001 From: LELlierle <2958542913@qq.com> Date: Fri, 19 Dec 2025 14:35:25 +0800 Subject: [PATCH 1/4] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E9=80=89=E9=A2=98?= =?UTF-8?q?=E5=92=8CREADME=E6=96=87=E4=BB=B6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/mmap_drl_nav/README.md | 61 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 61 insertions(+) create mode 100644 src/mmap_drl_nav/README.md diff --git a/src/mmap_drl_nav/README.md b/src/mmap_drl_nav/README.md new file mode 100644 index 0000000000..b24190554a --- /dev/null +++ b/src/mmap_drl_nav/README.md @@ -0,0 +1,61 @@ +# 机器人导航系统 - 环境配置与运行手册 +## 一、环境设置 +运行本项目代码前,需确保开发环境满足以下依赖版本要求: +- Python 3.7 及以上版本 +- PyTorch 1.7 及以上版本 +- CARLA 模拟器 0.9.11 版本 +- 其余 Python 依赖项详见项目根目录下的 `requirements.txt` 文件 + +## 二、依赖安装步骤 +### 1. 克隆项目仓库 +```bash +git clone https://github.com/yourusername/robot_navigation_system.git +cd robot_navigation_system +``` + +### 2. 安装Python依赖 +```bash +pip install -r requirements.txt +``` + +> **重要提示**:需提前完成CARLA模拟器的安装,并正确配置相关环境变量。CARLA安装与配置可参考官方文档:[https://carla.readthedocs.io/en/latest/build_linux/](https://carla.readthedocs.io/en/latest/build_linux/) + +## 三、代码运行方法 +### 运行系统模拟 +执行以下命令可启动CARLA模拟器,加载完整的感知、注意力与决策模块,生成机器人导航策略: +```bash +python run_simulation.py +``` + +## 四、模型训练与测试 +### 1. 训练模型 +执行以下命令加载模拟数据集,训练感知、跨域注意力和决策模块: +```bash +python main.py --mode train +``` +> 说明:训练参数、超参数等可在 `main.py` 文件中按需配置。 + +### 2. 测试模型 +模型训练完成后,执行以下命令加载测试数据集,评估模型在新数据上的表现并输出测试损失: +```bash +python main.py --mode test +``` + +## 五、模型部署 +> 说明:智能车辆底层控制配置将后续发布,以下仅提供算法部署与推理的核心步骤。 + +### 1. Jetson_robot 模型训练与导出 +推理前需先训练模型,并将其导出为Jetson机器人兼容的ONNX格式: +```bash +python /_agent/_lightweight/train.py +``` + +### 2. 加载模型并推理 +```bash +python _jetson_robot/deploy.py --onnx_model_path /_model/mmap_model.onnx +``` + +### 总结 +1. 环境核心要求:Python 3.7+、PyTorch 1.7+、CARLA 0.9.11,需确保CARLA环境变量配置正确; +2. 核心操作命令:模拟运行用 `run_simulation.py`,模型训练/测试用 `main.py --mode [train/test]`; +3. Jetson部署关键步骤:先训练导出ONNX模型,再通过 `deploy.py` 指定模型路径加载推理。 \ No newline at end of file From 9289619176e9e6c0fb792cb0b7443581c0342821 Mon Sep 17 00:00:00 2001 From: LELlierle <2958542913@qq.com> Date: Fri, 19 Dec 2025 15:53:06 +0800 Subject: [PATCH 2/4] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E9=80=89=E9=A2=98?= =?UTF-8?q?=E5=92=8CREADME=E6=96=87=E4=BB=B6?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/mmap_drl_nav/README.md | 25 +--- src/mmap_drl_nav/main.py | 59 ++++++++ src/mmap_drl_nav/run_simulation.py | 215 +++++++++++++++++++++++++++++ 3 files changed, 275 insertions(+), 24 deletions(-) create mode 100644 src/mmap_drl_nav/main.py create mode 100644 src/mmap_drl_nav/run_simulation.py diff --git a/src/mmap_drl_nav/README.md b/src/mmap_drl_nav/README.md index b24190554a..3ab8967222 100644 --- a/src/mmap_drl_nav/README.md +++ b/src/mmap_drl_nav/README.md @@ -7,19 +7,10 @@ - 其余 Python 依赖项详见项目根目录下的 `requirements.txt` 文件 ## 二、依赖安装步骤 -### 1. 克隆项目仓库 -```bash -git clone https://github.com/yourusername/robot_navigation_system.git -cd robot_navigation_system -``` - -### 2. 安装Python依赖 +### 1. 安装Python依赖 ```bash pip install -r requirements.txt ``` - -> **重要提示**:需提前完成CARLA模拟器的安装,并正确配置相关环境变量。CARLA安装与配置可参考官方文档:[https://carla.readthedocs.io/en/latest/build_linux/](https://carla.readthedocs.io/en/latest/build_linux/) - ## 三、代码运行方法 ### 运行系统模拟 执行以下命令可启动CARLA模拟器,加载完整的感知、注意力与决策模块,生成机器人导航策略: @@ -41,20 +32,6 @@ python main.py --mode train python main.py --mode test ``` -## 五、模型部署 -> 说明:智能车辆底层控制配置将后续发布,以下仅提供算法部署与推理的核心步骤。 - -### 1. Jetson_robot 模型训练与导出 -推理前需先训练模型,并将其导出为Jetson机器人兼容的ONNX格式: -```bash -python /_agent/_lightweight/train.py -``` - -### 2. 加载模型并推理 -```bash -python _jetson_robot/deploy.py --onnx_model_path /_model/mmap_model.onnx -``` - ### 总结 1. 环境核心要求:Python 3.7+、PyTorch 1.7+、CARLA 0.9.11,需确保CARLA环境变量配置正确; 2. 核心操作命令:模拟运行用 `run_simulation.py`,模型训练/测试用 `main.py --mode [train/test]`; diff --git a/src/mmap_drl_nav/main.py b/src/mmap_drl_nav/main.py new file mode 100644 index 0000000000..3756dc661d --- /dev/null +++ b/src/mmap_drl_nav/main.py @@ -0,0 +1,59 @@ +import torch +import torch.nn as nn +from torch.optim import Adam +from utils.dataloader import get_dataloader +from models.perception_module import PerceptionModule +from models.attention_module import CrossDomainAttention +from models.decision_module import DecisionModule +from models.sagm import SelfAssessmentGradientModel + +class IntegratedSystem: + def __init__(self, device='cpu', state_dim=128, action_dim=2): + self.device = device + self.perception = PerceptionModule().to(self.device) + self.attention = CrossDomainAttention(num_blocks=6).to(self.device) + self.decision = DecisionModule().to(self.device) + self.sagm = SelfAssessmentGradientModel(input_dim=state_dim + action_dim).to(self.device) + + def forward(self, image, lidar_data, imu_data, action): + scene_info, segmentation, odometry, obstacles, boundary = self.perception(imu_data, image, lidar_data) + fused_features = self.attention(scene_info, segmentation, odometry, obstacles, boundary) + + policy, value = self.decision(fused_features) + + sagm_q_value = self.sagm(fused_features, action) + return policy, value, sagm_q_value + +def train_model(model, dataloader, optimizer, device, num_epochs=10): + model.train() + for epoch in range(num_epochs): + running_loss = 0.0 + for i, (image, lidar_data, imu_data, target_action) in enumerate(dataloader): + image, lidar_data, imu_data, target_action = image.to(device), lidar_data.to(device), imu_data.to(device), target_action.to(device) + optimizer.zero_grad() + policy_output, value_output, sagm_q_value = model(image, lidar_data, imu_data, target_action) + + loss = (nn.MSELoss()(policy_output, target_action) + + nn.MSELoss()(value_output, target_action.sum(dim=1, keepdim=True)) + + nn.MSELoss()(sagm_q_value, target_action.sum(dim=1, keepdim=True))) + loss.backward() + optimizer.step() + running_loss += loss.item() + if i % 10 == 9: + print(f'Epoch [{epoch+1}/{num_epochs}], Batch [{i+1}/{len(dataloader)}], Loss: {running_loss / 10:.4f}') + running_loss = 0.0 + print('Training complete') + +def test_model(model, dataloader, device): + model.eval() + total_loss = 0.0 + with torch.no_grad(): + for image, lidar_data, imu_data, target_action in dataloader: + image, lidar_data, imu_data, target_action = image.to(device), lidar_data.to(device), imu_data.to(device), target_action.to(device) + policy_output, value_output, sagm_q_value = model(image, lidar_data, imu_data, target_action) + loss = (nn.MSELoss()(policy_output, target_action) + + nn.MSELoss()(value_output, target_action.sum(dim=1, keepdim=True)) + + nn.MSELoss()(sagm_q_value, target_action.sum(dim=1, keepdim=True))) + total_loss += loss.item() + avg_loss = total_loss / len(dataloader) + print(f'Test Average Loss: {avg_loss:.4f}') diff --git a/src/mmap_drl_nav/run_simulation.py b/src/mmap_drl_nav/run_simulation.py new file mode 100644 index 0000000000..f5d811617d --- /dev/null +++ b/src/mmap_drl_nav/run_simulation.py @@ -0,0 +1,215 @@ +import torch +import time +import logging +import carla +from typing import Tuple, Optional +from models.perception_module import PerceptionModule +from models.attention_module import CrossDomainAttention +from models.decision_module import DecisionModule +from models.dqn_agent import DQNAgent +from envs.carla_environment import CarlaEnvironment + +# 配置日志 +logging.basicConfig( + level=logging.INFO, + format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', + handlers=[ + logging.StreamHandler() + ] +) +logger = logging.getLogger(__name__) + +# 系统配置 +CONFIG = { + "simulation_steps": 100, + "step_delay": 0.1, + "carla_host": "localhost", + "carla_port": 2000, + "image_size": (3, 256, 256), + "lidar_size": (1, 256, 256), + "imu_size": (1, 6), + "max_throttle": 1.0, + "max_steer": 1.0 +} + + +class IntegratedSystem: + """ + 自动驾驶集成系统,整合感知、注意力和决策模块 + """ + + def __init__(self, device: str = 'cpu'): + self.device = torch.device(device) + self.logger = logging.getLogger(self.__class__.__name__) + + # 初始化各个模块 + try: + self.perception = PerceptionModule().to(self.device) + self.attention = CrossDomainAttention(num_blocks=6).to(self.device) + self.decision = DecisionModule().to(self.device) + self.logger.info(f"成功初始化所有模块,使用设备: {self.device}") + except Exception as e: + self.logger.error(f"模块初始化失败: {e}") + raise + + def forward(self, image: torch.Tensor, lidar_data: torch.Tensor, imu_data: torch.Tensor) -> Tuple[ + torch.Tensor, torch.Tensor]: + """ + 前向传播,处理输入数据并输出策略和价值 + + Args: + image: 图像数据 [B, 3, H, W] + lidar_data: LiDAR数据 [B, 1, H, W] + imu_data: IMU数据 [B, 6] + + Returns: + policy: 决策策略 [B, num_actions] + value: 状态价值 [B, 1] + """ + try: + # 感知模块处理 + scene_info, segmentation, odometry, obstacles, boundary = self.perception(imu_data, image, lidar_data) + + # 跨域注意力融合特征 + fused_features = self.attention(scene_info, segmentation, odometry, obstacles, boundary) + + # 决策模块输出 + policy, value = self.decision(fused_features) + + return policy, value + except Exception as e: + self.logger.error(f"前向传播过程出错: {e}") + raise + + +def init_carla_environment(host: str = "localhost", port: int = 2000) -> Optional[CarlaEnvironment]: + """ + 初始化CARLA环境 + + Args: + host: CARLA服务器地址 + port: CARLA服务器端口 + + Returns: + 初始化后的CarlaEnvironment实例 + """ + try: + env = CarlaEnvironment(host=host, port=port) + # 等待环境加载完成 + time.sleep(2.0) + logger.info(f"成功连接到CARLA服务器 {host}:{port}") + + # 检查车辆是否正确初始化 + if hasattr(env, 'vehicle') and env.vehicle is not None: + logger.info("CARLA车辆已成功初始化") + else: + logger.error("CARLA车辆初始化失败") + return None + + return env + except Exception as e: + logger.error(f"CARLA环境初始化失败: {e}") + return None + + +def clamp_control_values(throttle: float, steer: float) -> Tuple[float, float]: + """ + 限制控制信号在合法范围内 + + Args: + throttle: 油门值 + steer: 转向值 + + Returns: + 限制后的油门和转向值 + """ + # 限制油门在 0-1 之间 + throttle_clamped = max(0.0, min(CONFIG["max_throttle"], throttle)) + # 限制转向在 -1 到 1 之间 + steer_clamped = max(-CONFIG["max_steer"], min(CONFIG["max_steer"], steer)) + + return throttle_clamped, steer_clamped + + +def run_simulation(): + """运行CARLA仿真主函数""" + # 1. 初始化CARLA环境 + env = init_carla_environment( + host=CONFIG["carla_host"], + port=CONFIG["carla_port"] + ) + if env is None: + logger.error("无法初始化CARLA环境,退出仿真") + return + + # 2. 初始化集成系统 + try: + device = 'cuda' if torch.cuda.is_available() else 'cpu' + system = IntegratedSystem(device=device) + logger.info(f"集成系统初始化完成,使用设备: {device}") + except Exception as e: + logger.error(f"集成系统初始化失败: {e}") + return + + # 3. 运行仿真步骤 + logger.info(f"开始运行{CONFIG['simulation_steps']}步仿真") + for step in range(CONFIG["simulation_steps"]): + try: + # 生成模拟输入数据(实际应用中应从CARLA环境获取真实数据) + image = torch.randn(*CONFIG["image_size"]).unsqueeze(0).to(system.device) + lidar_data = torch.randn(*CONFIG["lidar_size"]).unsqueeze(0).to(system.device) + imu_data = torch.randn(*CONFIG["imu_size"]).to(system.device) + + # 前向传播获取决策 + policy, value = system.forward(image, lidar_data, imu_data) + + # 转换并限制控制信号 + throttle, steer = clamp_control_values( + float(policy[0][0]), + float(policy[0][1]) + ) + + # 应用控制信号到车辆 + control = carla.VehicleControl(throttle=throttle, steer=steer) + env.vehicle.apply_control(control) + + # 日志输出 + if step % 10 == 0: # 每10步输出一次日志 + logger.info( + f"仿真步骤 {step:3d} | 油门: {throttle:.3f} | " + f"转向: {steer:.3f} | 状态价值: {value[0][0]:.3f}" + ) + + # 模拟步骤间延迟 + time.sleep(CONFIG["step_delay"]) + + except Exception as e: + logger.error(f"仿真步骤 {step} 出错: {e}") + continue + + # 4. 清理资源 + try: + # 停止车辆 + env.vehicle.apply_control(carla.VehicleControl(throttle=0.0, brake=1.0)) + time.sleep(1.0) + + # 清理CARLA环境 + if hasattr(env, 'cleanup'): + env.cleanup() + + logger.info("仿真完成,已清理所有资源") + except Exception as e: + logger.error(f"清理资源时出错: {e}") + + +if __name__ == "__main__": + # 设置随机种子以保证可复现性 + torch.manual_seed(42) + + # 运行仿真 + try: + run_simulation() + except KeyboardInterrupt: + logger.info("用户中断了仿真") + except Exception as e: + logger.error(f"仿真运行出错: {e}") \ No newline at end of file From 12476158b57bf588a0d4c0541dae18ca6f9a9a95 Mon Sep 17 00:00:00 2001 From: LELlierle <2958542913@qq.com> Date: Fri, 19 Dec 2025 17:12:17 +0800 Subject: [PATCH 3/4] =?UTF-8?q?=E6=8F=90=E4=BA=A4=E9=80=89=E9=A2=98?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/mmap_drl_nav/README.md | 39 +----- src/mmap_drl_nav/main.py | 59 -------- src/mmap_drl_nav/run_simulation.py | 215 ----------------------------- 3 files changed, 1 insertion(+), 312 deletions(-) delete mode 100644 src/mmap_drl_nav/main.py delete mode 100644 src/mmap_drl_nav/run_simulation.py diff --git a/src/mmap_drl_nav/README.md b/src/mmap_drl_nav/README.md index 3ab8967222..d22090f08c 100644 --- a/src/mmap_drl_nav/README.md +++ b/src/mmap_drl_nav/README.md @@ -1,38 +1 @@ -# 机器人导航系统 - 环境配置与运行手册 -## 一、环境设置 -运行本项目代码前,需确保开发环境满足以下依赖版本要求: -- Python 3.7 及以上版本 -- PyTorch 1.7 及以上版本 -- CARLA 模拟器 0.9.11 版本 -- 其余 Python 依赖项详见项目根目录下的 `requirements.txt` 文件 - -## 二、依赖安装步骤 -### 1. 安装Python依赖 -```bash -pip install -r requirements.txt -``` -## 三、代码运行方法 -### 运行系统模拟 -执行以下命令可启动CARLA模拟器,加载完整的感知、注意力与决策模块,生成机器人导航策略: -```bash -python run_simulation.py -``` - -## 四、模型训练与测试 -### 1. 训练模型 -执行以下命令加载模拟数据集,训练感知、跨域注意力和决策模块: -```bash -python main.py --mode train -``` -> 说明:训练参数、超参数等可在 `main.py` 文件中按需配置。 - -### 2. 测试模型 -模型训练完成后,执行以下命令加载测试数据集,评估模型在新数据上的表现并输出测试损失: -```bash -python main.py --mode test -``` - -### 总结 -1. 环境核心要求:Python 3.7+、PyTorch 1.7+、CARLA 0.9.11,需确保CARLA环境变量配置正确; -2. 核心操作命令:模拟运行用 `run_simulation.py`,模型训练/测试用 `main.py --mode [train/test]`; -3. Jetson部署关键步骤:先训练导出ONNX模型,再通过 `deploy.py` 指定模型路径加载推理。 \ No newline at end of file +我的选题是机器人导航系统 diff --git a/src/mmap_drl_nav/main.py b/src/mmap_drl_nav/main.py deleted file mode 100644 index 3756dc661d..0000000000 --- a/src/mmap_drl_nav/main.py +++ /dev/null @@ -1,59 +0,0 @@ -import torch -import torch.nn as nn -from torch.optim import Adam -from utils.dataloader import get_dataloader -from models.perception_module import PerceptionModule -from models.attention_module import CrossDomainAttention -from models.decision_module import DecisionModule -from models.sagm import SelfAssessmentGradientModel - -class IntegratedSystem: - def __init__(self, device='cpu', state_dim=128, action_dim=2): - self.device = device - self.perception = PerceptionModule().to(self.device) - self.attention = CrossDomainAttention(num_blocks=6).to(self.device) - self.decision = DecisionModule().to(self.device) - self.sagm = SelfAssessmentGradientModel(input_dim=state_dim + action_dim).to(self.device) - - def forward(self, image, lidar_data, imu_data, action): - scene_info, segmentation, odometry, obstacles, boundary = self.perception(imu_data, image, lidar_data) - fused_features = self.attention(scene_info, segmentation, odometry, obstacles, boundary) - - policy, value = self.decision(fused_features) - - sagm_q_value = self.sagm(fused_features, action) - return policy, value, sagm_q_value - -def train_model(model, dataloader, optimizer, device, num_epochs=10): - model.train() - for epoch in range(num_epochs): - running_loss = 0.0 - for i, (image, lidar_data, imu_data, target_action) in enumerate(dataloader): - image, lidar_data, imu_data, target_action = image.to(device), lidar_data.to(device), imu_data.to(device), target_action.to(device) - optimizer.zero_grad() - policy_output, value_output, sagm_q_value = model(image, lidar_data, imu_data, target_action) - - loss = (nn.MSELoss()(policy_output, target_action) + - nn.MSELoss()(value_output, target_action.sum(dim=1, keepdim=True)) + - nn.MSELoss()(sagm_q_value, target_action.sum(dim=1, keepdim=True))) - loss.backward() - optimizer.step() - running_loss += loss.item() - if i % 10 == 9: - print(f'Epoch [{epoch+1}/{num_epochs}], Batch [{i+1}/{len(dataloader)}], Loss: {running_loss / 10:.4f}') - running_loss = 0.0 - print('Training complete') - -def test_model(model, dataloader, device): - model.eval() - total_loss = 0.0 - with torch.no_grad(): - for image, lidar_data, imu_data, target_action in dataloader: - image, lidar_data, imu_data, target_action = image.to(device), lidar_data.to(device), imu_data.to(device), target_action.to(device) - policy_output, value_output, sagm_q_value = model(image, lidar_data, imu_data, target_action) - loss = (nn.MSELoss()(policy_output, target_action) + - nn.MSELoss()(value_output, target_action.sum(dim=1, keepdim=True)) + - nn.MSELoss()(sagm_q_value, target_action.sum(dim=1, keepdim=True))) - total_loss += loss.item() - avg_loss = total_loss / len(dataloader) - print(f'Test Average Loss: {avg_loss:.4f}') diff --git a/src/mmap_drl_nav/run_simulation.py b/src/mmap_drl_nav/run_simulation.py deleted file mode 100644 index f5d811617d..0000000000 --- a/src/mmap_drl_nav/run_simulation.py +++ /dev/null @@ -1,215 +0,0 @@ -import torch -import time -import logging -import carla -from typing import Tuple, Optional -from models.perception_module import PerceptionModule -from models.attention_module import CrossDomainAttention -from models.decision_module import DecisionModule -from models.dqn_agent import DQNAgent -from envs.carla_environment import CarlaEnvironment - -# 配置日志 -logging.basicConfig( - level=logging.INFO, - format='%(asctime)s - %(name)s - %(levelname)s - %(message)s', - handlers=[ - logging.StreamHandler() - ] -) -logger = logging.getLogger(__name__) - -# 系统配置 -CONFIG = { - "simulation_steps": 100, - "step_delay": 0.1, - "carla_host": "localhost", - "carla_port": 2000, - "image_size": (3, 256, 256), - "lidar_size": (1, 256, 256), - "imu_size": (1, 6), - "max_throttle": 1.0, - "max_steer": 1.0 -} - - -class IntegratedSystem: - """ - 自动驾驶集成系统,整合感知、注意力和决策模块 - """ - - def __init__(self, device: str = 'cpu'): - self.device = torch.device(device) - self.logger = logging.getLogger(self.__class__.__name__) - - # 初始化各个模块 - try: - self.perception = PerceptionModule().to(self.device) - self.attention = CrossDomainAttention(num_blocks=6).to(self.device) - self.decision = DecisionModule().to(self.device) - self.logger.info(f"成功初始化所有模块,使用设备: {self.device}") - except Exception as e: - self.logger.error(f"模块初始化失败: {e}") - raise - - def forward(self, image: torch.Tensor, lidar_data: torch.Tensor, imu_data: torch.Tensor) -> Tuple[ - torch.Tensor, torch.Tensor]: - """ - 前向传播,处理输入数据并输出策略和价值 - - Args: - image: 图像数据 [B, 3, H, W] - lidar_data: LiDAR数据 [B, 1, H, W] - imu_data: IMU数据 [B, 6] - - Returns: - policy: 决策策略 [B, num_actions] - value: 状态价值 [B, 1] - """ - try: - # 感知模块处理 - scene_info, segmentation, odometry, obstacles, boundary = self.perception(imu_data, image, lidar_data) - - # 跨域注意力融合特征 - fused_features = self.attention(scene_info, segmentation, odometry, obstacles, boundary) - - # 决策模块输出 - policy, value = self.decision(fused_features) - - return policy, value - except Exception as e: - self.logger.error(f"前向传播过程出错: {e}") - raise - - -def init_carla_environment(host: str = "localhost", port: int = 2000) -> Optional[CarlaEnvironment]: - """ - 初始化CARLA环境 - - Args: - host: CARLA服务器地址 - port: CARLA服务器端口 - - Returns: - 初始化后的CarlaEnvironment实例 - """ - try: - env = CarlaEnvironment(host=host, port=port) - # 等待环境加载完成 - time.sleep(2.0) - logger.info(f"成功连接到CARLA服务器 {host}:{port}") - - # 检查车辆是否正确初始化 - if hasattr(env, 'vehicle') and env.vehicle is not None: - logger.info("CARLA车辆已成功初始化") - else: - logger.error("CARLA车辆初始化失败") - return None - - return env - except Exception as e: - logger.error(f"CARLA环境初始化失败: {e}") - return None - - -def clamp_control_values(throttle: float, steer: float) -> Tuple[float, float]: - """ - 限制控制信号在合法范围内 - - Args: - throttle: 油门值 - steer: 转向值 - - Returns: - 限制后的油门和转向值 - """ - # 限制油门在 0-1 之间 - throttle_clamped = max(0.0, min(CONFIG["max_throttle"], throttle)) - # 限制转向在 -1 到 1 之间 - steer_clamped = max(-CONFIG["max_steer"], min(CONFIG["max_steer"], steer)) - - return throttle_clamped, steer_clamped - - -def run_simulation(): - """运行CARLA仿真主函数""" - # 1. 初始化CARLA环境 - env = init_carla_environment( - host=CONFIG["carla_host"], - port=CONFIG["carla_port"] - ) - if env is None: - logger.error("无法初始化CARLA环境,退出仿真") - return - - # 2. 初始化集成系统 - try: - device = 'cuda' if torch.cuda.is_available() else 'cpu' - system = IntegratedSystem(device=device) - logger.info(f"集成系统初始化完成,使用设备: {device}") - except Exception as e: - logger.error(f"集成系统初始化失败: {e}") - return - - # 3. 运行仿真步骤 - logger.info(f"开始运行{CONFIG['simulation_steps']}步仿真") - for step in range(CONFIG["simulation_steps"]): - try: - # 生成模拟输入数据(实际应用中应从CARLA环境获取真实数据) - image = torch.randn(*CONFIG["image_size"]).unsqueeze(0).to(system.device) - lidar_data = torch.randn(*CONFIG["lidar_size"]).unsqueeze(0).to(system.device) - imu_data = torch.randn(*CONFIG["imu_size"]).to(system.device) - - # 前向传播获取决策 - policy, value = system.forward(image, lidar_data, imu_data) - - # 转换并限制控制信号 - throttle, steer = clamp_control_values( - float(policy[0][0]), - float(policy[0][1]) - ) - - # 应用控制信号到车辆 - control = carla.VehicleControl(throttle=throttle, steer=steer) - env.vehicle.apply_control(control) - - # 日志输出 - if step % 10 == 0: # 每10步输出一次日志 - logger.info( - f"仿真步骤 {step:3d} | 油门: {throttle:.3f} | " - f"转向: {steer:.3f} | 状态价值: {value[0][0]:.3f}" - ) - - # 模拟步骤间延迟 - time.sleep(CONFIG["step_delay"]) - - except Exception as e: - logger.error(f"仿真步骤 {step} 出错: {e}") - continue - - # 4. 清理资源 - try: - # 停止车辆 - env.vehicle.apply_control(carla.VehicleControl(throttle=0.0, brake=1.0)) - time.sleep(1.0) - - # 清理CARLA环境 - if hasattr(env, 'cleanup'): - env.cleanup() - - logger.info("仿真完成,已清理所有资源") - except Exception as e: - logger.error(f"清理资源时出错: {e}") - - -if __name__ == "__main__": - # 设置随机种子以保证可复现性 - torch.manual_seed(42) - - # 运行仿真 - try: - run_simulation() - except KeyboardInterrupt: - logger.info("用户中断了仿真") - except Exception as e: - logger.error(f"仿真运行出错: {e}") \ No newline at end of file From 48921ee2cb8151c69f9e289ad0b7d2fa7039f582 Mon Sep 17 00:00:00 2001 From: LELlierle <2958542913@qq.com> Date: Sat, 20 Dec 2025 17:18:35 +0800 Subject: [PATCH 4/4] =?UTF-8?q?=E6=9B=B4=E6=96=B0READMA=E6=96=87=E4=BB=B6?= =?UTF-8?q?=EF=BC=8C=E9=85=8D=E7=BD=AE=E7=8E=AF=E5=A2=83?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/mmap_drl_nav/README.md | 43 +++++++++++++++++++++++++++++++++++++- 1 file changed, 42 insertions(+), 1 deletion(-) diff --git a/src/mmap_drl_nav/README.md b/src/mmap_drl_nav/README.md index d22090f08c..bd8e78c6ad 100644 --- a/src/mmap_drl_nav/README.md +++ b/src/mmap_drl_nav/README.md @@ -1 +1,42 @@ -我的选题是机器人导航系统 +# 多模式机器人导航系统 +## 项目概述 +你要开发的多模式机器人导航系统,核心是融合IMU、摄像头、激光雷达等多传感器数据,通过感知、跨域注意力、决策三大模块协作,最终输出机器人的行动策略,该系统基于CARLA 0.9.11模拟器开发,并采用深度学习技术完成训练与测试。 + +## 环境设置 +### 1. 基础环境要求 +- Python版本:3.7及以上 +- PyTorch版本:1.7及以上(建议匹配CUDA版本以提升训练效率) +- CARLA模拟器版本:0.9.11(需保证客户端与服务端版本一致) + +### 2. 快速配置步骤 +```bash +# 1. 创建并激活虚拟环境 +conda create -n robot_nav python=3.7 -y +conda activate robot_nav + +# 2. 安装PyTorch(以CUDA 11.0为例,可按需调整) +pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 -f https://download.pytorch.org/whl/torch_stable.html + +# 3. 安装CARLA 0.9.11 +pip install carla==0.9.11 + +# 4. 补充常用依赖 +pip install numpy opencv-python scipy matplotlib +``` + +### 3. 环境验证 +```bash +# 验证Python版本 +python --version + +# 验证PyTorch +python -c "import torch; print(f'PyTorch版本:{torch.__version__},CUDA可用:{torch.cuda.is_available()}')" + +# 验证CARLA(需先启动CARLA 0.9.11服务端) +python -c "import carla; client=carla.Client('localhost',2000); client.set_timeout(5.0); print('CARLA连接成功' if client.get_world() else 'CARLA连接失败')" +``` + +## 总结 +1. 环境核心要求:Python 3.7+、PyTorch 1.7+、CARLA 0.9.11,版本匹配是关键。 +2. 配置流程:先创建虚拟环境,再依次安装PyTorch、CARLA及辅助依赖,最后验证环境可用性。 +3. 系统核心逻辑:通过三大模块融合多传感器数据,基于CARLA模拟器实现机器人导航策略输出。 \ No newline at end of file