From 393d0487edc690d7f3e63e023fbacb040d146e6b Mon Sep 17 00:00:00 2001 From: lan Date: Thu, 18 Dec 2025 18:44:05 +0800 Subject: [PATCH] =?UTF-8?q?=E6=B7=BB=E5=8A=A0=E4=B8=A4=E4=B8=AA=E9=9D=9E?= =?UTF-8?q?=E5=B8=B8=E9=87=8D=E8=A6=81=E7=9A=84=E6=B3=A8=E9=87=8A?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- src/Neuro_Mujoco/main.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/Neuro_Mujoco/main.py b/src/Neuro_Mujoco/main.py index 0f9d25aa49..890b1ee711 100644 --- a/src/Neuro_Mujoco/main.py +++ b/src/Neuro_Mujoco/main.py @@ -307,7 +307,7 @@ def control_cmd_callback(msg: Float32MultiArray): obs_tensor = torch.zeros(1, obs_dimension, dtype=torch.float32) if policy_net else None while viewer_instance.is_running() and (not use_ros_mode or not rospy.is_shutdown()): - # 控制指令优先级:ROS指令 > 策略推理 > 无控制 + # 【核心优先级规则】控制指令优先级:ROS外部指令 > 策略网络推理 > 无控制输入 if use_ros_mode and ros_ctrl_cmd is not None: data.ctrl[:] = ros_ctrl_cmd elif policy_net is not None: @@ -323,7 +323,7 @@ def control_cmd_callback(msg: Float32MultiArray): # 策略推理得到归一化动作([-1,1]) normalized_action = policy_net(obs_tensor).squeeze().numpy() - # 映射到执行器实际控制范围 + 裁剪(核心优化:删除重复映射) + # 【关键安全机制】线性映射归一化动作到执行器物理极限,并强制裁剪防止超限损坏模拟 if control_range is not None: # 线性映射:[-1,1] → [ctrl_min, ctrl_max] normalized_action = control_range[:, 0] + (control_range[:, 1] - control_range[:, 0]) * (normalized_action + 1) / 2