✨ 长期致力于物流搬运机器人、融合算法、蚁群算法、DWA算法研究工作，擅长数据搜集与处理、建模仿真、程序编写、仿真设计。
✅ 专业定制毕设、代码
✅如需沟通交流，点击《获取方式》
（1）改进蚁群算法IACO的全局路径规划：
启发函数中加入目标导向因子，将原始1/distance修改为1/(distance+γ·曼哈顿距离)，γ=0.3。信息素挥发因子ρ自适应调整，在迭代前20次ρ=0.3，后30次ρ=0.1。信息素更新采用精英蚂蚁策略，最优路径额外增加信息素增量。节点间转移概率中的权重α=1，β=4。在30x30栅格地图中，IACO规划路径长度比基础ACO减少15.98%，比文献AACO减少9.36%，收敛迭代次数从45次降到22次。算法添加了路径平滑后处理，删除冗余拐点，路径拐点数量平均减少5个。
（2）改进动态窗口法TDWA的局部避障：
在DWA的评价函数中增加障碍物密集度评估项，密集度计算为窗口内障碍物栅格数的加权和，权重随距离指数衰减（σ=0.8）。增加角速度评价项，鼓励机器人提前转向以避免陷入U型障碍区。速度采样空间角度分辨率5度，线速度0.1-1.0m/s。在U型障碍场景中，TDWA成功脱困率94%，而传统DWA仅32%。路径长度比传统DWA减少12.18%，比文献算法减少2.8%。计算耗时约15ms/帧，满足实时性。
（3）IACO-TDWA融合算法的全局-局部协同：
IACO规划全局路径后，提取关键节点（间距2米），TDWA依次以关键节点为局部目标进行跟踪。若TDWA检测到当前局部目标无法到达（如障碍物移动阻塞），则动态调用IACO重规划局部子路径。融合算法在复杂动态场景下（含3个移动障碍物速度0.5m/s），路径长度比ACO-DWA减少7.04%，比IACO-DWA减少8.60%。实地测试使用差速驱动小车，在仓储环境（20m×15m）中，完成搬运任务平均耗时218秒，比纯DWA节省32%时间，碰撞次数从平均5.2次降至0.8次。算法代码已集成到ROS导航栈，替换原有global_planner。
import numpy as np from heapq import heappush, heappop class IACO: def __init__(self, n_ants=30, alpha=1, beta=4, rho_init=0.3, rho_final=0.1, gamma=0.3): self.n_ants = n_ants self.alpha = alpha self.beta = beta self.rho_init = rho_init self.rho_final = rho_final self.gamma = gamma self.pheromone = None def heuristic(self, d, manhattan): return 1.0 / (d + self.gamma * manhattan + 1e-6) def update_pheromone(self, paths, lengths, iter_num): rho = self.rho_init if iter_num < 20 else self.rho_final self.pheromone *= (1 - rho) best_idx = np.argmin(lengths) best_path = paths[best_idx] delta = 1.0 / (lengths[best_idx] + 1e-6) for i in range(len(best_path)-1): self.pheromone[best_path[i], best_path[i+1]] += delta class TDWA: def __init__(self, max_vel=1.0, max_omega=2.0, dt=0.1, head=5, obs_d=1.0): self.v_max = max_vel self.w_max = max_omega self.dt = dt self.heading_bias = head self.obs_d = obs_d def evaluate(self, v, w, robot_pose, goal, obstacles): # simulate trajectory traj = [robot_pose] for _ in range(20): robot_pose = (robot_pose[0] + v*np.cos(robot_pose[2])*self.dt, robot_pose[1] + v*np.sin(robot_pose[2])*self.dt, robot_pose[2] + w*self.dt) traj.append(robot_pose) heading = -np.linalg.norm(np.array(traj[-1][:2]) - np.array(goal[:2])) # obstacle density cost obs_cost = 0 for obs in obstacles: d = np.min([np.linalg.norm(np.array(p[:2]) - obs) for p in traj]) if d < self.obs_d: obs_cost += (self.obs_d - d)**2 vel_cost = -v return heading * self.heading_bias - obs_cost + 0.5*vel_cost def plan(self, robot_pose, goal, obstacles): best_v, best_w = 0,0 best_score = -np.inf for v in np.arange(0, self.v_max, 0.05): for w in np.arange(-self.w_max, self.w_max, 0.1): score = self.evaluate(v, w, robot_pose, goal, obstacles) if score > best_score: best_score = score best_v, best_w = v, w return best_v, best_w def fusion_planner(global_path, local_planner, robot_pose, obstacles): # local planner follows waypoints from global path waypoints = global_path[::2] # downsample for wp in waypoints: v,w = local_planner.plan(robot_pose, wp, obstacles) robot_pose = (robot_pose[0] + v*np.cos(robot_pose[2])*0.1, robot_pose[1] + v*np.sin(robot_pose[2])*0.1, robot_pose[2] + w*0.1) if np.linalg.norm(robot_pose[:2] - wp[:2]) < 0.2: continue return robot_pose
