From 0b05c010244ac3bad67612a81f840aab4f3f9cf4 Mon Sep 17 00:00:00 2001 From: gouhanke <12219217+gouhanke@user.noreply.gitee.com> Date: Thu, 12 Feb 2026 19:54:11 +0800 Subject: [PATCH] =?UTF-8?q?feat:=20=E6=8E=A8=E7=90=86=E6=97=B6=E8=BE=93?= =?UTF-8?q?=E5=87=BAaction?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- roboimi/demos/vla_scripts/eval_vla.py | 7 +++++++ roboimi/vla/conf/agent/resnet_diffusion.yaml | 4 ++-- roboimi/vla/conf/eval/eval.yaml | 5 +++++ 3 files changed, 14 insertions(+), 2 deletions(-) diff --git a/roboimi/demos/vla_scripts/eval_vla.py b/roboimi/demos/vla_scripts/eval_vla.py index 9c358e4..6b967ed 100644 --- a/roboimi/demos/vla_scripts/eval_vla.py +++ b/roboimi/demos/vla_scripts/eval_vla.py @@ -243,6 +243,13 @@ def main(cfg: DictConfig): # 转换为 numpy action = action.cpu().numpy() + # 调试:打印当前时间步的动作(由配置控制) + if eval_cfg.get('verbose_action', False): + print(f"\n[Step {t:3d}] 预测动作: {action}") + print(f" - 动作形状: {action.shape}") + print(f" - 动作范围: [{action.min():.4f}, {action.max():.4f}]") + print(f" - 动作均值: {action.mean():.4f}, 标准差: {action.std():.4f}") + # 可选:平滑动作 if smoother: action = smoother.smooth(action) diff --git a/roboimi/vla/conf/agent/resnet_diffusion.yaml b/roboimi/vla/conf/agent/resnet_diffusion.yaml index 3574f96..13c18a0 100644 --- a/roboimi/vla/conf/agent/resnet_diffusion.yaml +++ b/roboimi/vla/conf/agent/resnet_diffusion.yaml @@ -23,9 +23,9 @@ normalization_type: "min_max" # "min_max" or "gaussian" # ==================== # 时间步配置 # ==================== -pred_horizon: 16 # 预测未来多少步动作 +pred_horizon: 8 # 预测未来多少步动作 obs_horizon: 2 # 使用多少步历史观测 -num_action_steps: 8 # 每次推理实际执行多少步动作(应 <= pred_horizon - obs_horizon + 1) +num_action_steps: 4 # 每次推理实际执行多少步动作(应 <= pred_horizon - obs_horizon + 1) # ==================== # 相机配置 diff --git a/roboimi/vla/conf/eval/eval.yaml b/roboimi/vla/conf/eval/eval.yaml index 0b6f345..2960937 100644 --- a/roboimi/vla/conf/eval/eval.yaml +++ b/roboimi/vla/conf/eval/eval.yaml @@ -26,4 +26,9 @@ use_smoothing: false smooth_method: "ema" smooth_alpha: 0.3 +# ==================== +# 调试选项 +# ==================== +verbose_action: true # 是否打印每个时间步的动作信息 +