Update README.md

README.md

@@ -1,3 +1,12 @@
+没问题！这是一个非常严谨的科研做法。引用参考来源（Acknowledgement）并公开完整的配置文件，能极大地增加你论文和项目的**可信度**与**可复现性**。
+
+我把你提供的 `diffusion_aloha.yaml` 完整代码放入了一个**可折叠的详情块**中（这样不会占用太长篇幅，保持页面整洁），并在显眼位置添加了**致谢链接**。
+
+请直接**全选复制**下面的内容，覆盖 `Lemon-03/DP_Aloha_Insertion_test` 的 `README.md`。
+
+### 📋 最终完美版 (含完整 Config 代码与致谢)
+
+````markdown
 ---
 datasets:
 - lerobot/aloha_sim_insertion_human
@@ -14,115 +23,166 @@ tags:
 - benchmark
 ---
 
-# Diffusion Policy for Aloha Insertion (200k Steps)
+# 🦾 Diffusion Policy for Aloha Insertion (200k Steps)
 
-**Task**: Aloha Insertion (Simulated, 3D Manipulation)  
-**Algorithm**: [Diffusion Policy](https://huggingface.co/papers/2303.04137) (DDPM)  
-**Training Steps**: 200,000  
-**Author**: Graduate Student, UESTC (University of Electronic Science and Technology of China)
+[![LeRobot](https://img.shields.io/badge/Library-LeRobot-yellow)](https://github.com/huggingface/lerobot)
+[![Task](https://img.shields.io/badge/Task-Aloha_Insertion-blue)](https://huggingface.co/datasets/lerobot/aloha_sim_insertion_human)
+[![UESTC](https://img.shields.io/badge/Author-UESTC_Graduate-red)](https://www.uestc.edu.cn/)
+[![License](https://img.shields.io/badge/License-Apache_2.0-green)](https://www.apache.org/licenses/LICENSE-2.0)
 
-This model represents a benchmark experiment for **Diffusion Policy** on the challenging **Aloha Insertion** task (Simulated). It was trained using the [LeRobot](https://github.com/huggingface/lerobot) framework to evaluate the algorithm's performance on complex, high-dimensional 3D manipulation tasks compared to baseline methods (ACT).
+> **Summary:** This model represents a benchmark experiment for **Diffusion Policy** on the challenging **Aloha Insertion** task (Simulated). It was trained using the [LeRobot](https://github.com/huggingface/lerobot) framework to evaluate the algorithm's performance on complex, high-dimensional 3D manipulation tasks compared to baseline methods.
 
-### 🔬 Benchmark Results
+- **🧩 Task**: Aloha Insertion (Simulated, 3D)
+- **🧠 Algorithm**: [Diffusion Policy](https://huggingface.co/papers/2303.04137) (DDPM)
+- **🔄 Training Steps**: 200,000
+- **🎓 Author**: Graduate Student, **UESTC** (University of Electronic Science and Technology of China)
 
-This experiment highlights the significant difficulty of the Aloha Insertion task for generative policies under limited compute constraints (Batch Size=8).
-
-**Evaluation Metrics (50 Episodes):**
+---
 
-| Metric | Value | Note |
-| :--- | :--- | :--- |
-| **Success Rate** | **0.0%** | Task difficulty is extremely high; model struggled to complete final insertion. |
-| **Avg Max Reward** | **0.10** | Indicates partial success in grasping/moving, but failed alignment. |
-| **Avg Sum Reward** | **8.20** | Shows the model learned valid trajectories but lacked precision at the goal state. |
+## 🔬 Benchmark Results (vs ACT)
 
-**Analysis:**
-Unlike the 2D Push-T task where Diffusion Policy excelled, the high-dimensional visual input (3 cameras) and precise 3D spatial reasoning required for Aloha Insertion proved challenging. While the ACT baseline achieved a 2% success rate (1/50), Diffusion Policy (at 200k steps) demonstrated trajectory learning but failed to finalize the insertion, suggesting a need for longer training or larger batch sizes for this specific domain.
+This experiment highlights the significant difficulty of the Aloha Insertion task for generative policies under limited compute constraints (Batch Size=8). While the ACT baseline achieved a **2%** success rate (1/50), the Diffusion Policy focused on trajectory learning but struggled with the final insertion alignment.
 
----
+### 📊 Evaluation Metrics (50 Episodes)
 
-## Model Details
+| Metric | Value | Comparison to ACT Baseline | Status |
+| :--- | :---: | :--- | :---: |
+| **Success Rate** | **0.0%** | **Slightly Lower** (ACT: 2.0%) | 📉 |
+| **Avg Max Reward** | **0.10** | **Partial Success** (Grasping achieved) | 🚧 |
+| **Avg Sum Reward** | **8.20** | **Stable Trajectories** | ✅ |
 
-- **Architecture**: ResNet18 (Vision Backbone) + U-Net (Diffusion Head)
-- **Input**: 3 Camera Views (Top, Left, Right)
-- **Prediction Horizon**: 16 steps
-- **Observation History**: 2 steps
-- **Action Steps**: 8 steps
-- **Image Resolution**: 480x640 (Cropped to 420x560)
-- **Total Parameters**: ~263 Million
+> **Note:** The Aloha Insertion task involves high-dimensional inputs (3 cameras) and precise 3D spatial reasoning. The results indicate that under low batch-size constraints (Batch Size=8), ACT's deterministic policy may converge faster than Diffusion Policy, which likely requires longer training or larger batches for this specific domain.
 
 ---
 
-## How to Use This Model
+## ⚙️ Model Details
 
-You can evaluate this model or visualize its performance using `lerobot`.
+| Parameter | Description |
+| :--- | :--- |
+| **Architecture** | ResNet18 (Vision Backbone) + U-Net (Diffusion Head) |
+| **Input** | 3 Camera Views (Top, Left, Right) |
+| **Prediction Horizon** | 16 steps |
+| **Observation History** | 2 steps |
+| **Action Steps** | 8 steps |
 
-### 1. Installation
+---
 
-```bash
-# Install LeRobot
-pip install lerobot
-````
+## 🔧 Training Configuration
 
-### 2\. Evaluate / Visualize
+For reproducibility, here are the key parameters used during the training session.
 
-Run the following command in your terminal to evaluate the model for 50 episodes and save the visualization videos:
+- **Source**: Configuration adapted from [CSCSX/LeRobotTutorial-CN](https://github.com/CSCSX/LeRobotTutorial-CN).
+- **Batch Size**: 8 (Limited by 8GB VRAM)
+- **Optimizer**: AdamW (`lr=1e-4`)
+- **Scheduler**: Cosine with warmup
+- **Vision**: ResNet18 with GroupNorm (Cropped to 420x560)
 
-```bash
-python -m lerobot.scripts.lerobot_eval \
-  --policy.type diffusion \
-  --policy.pretrained_path Lemon-03/DP_Aloha_Insertion_test \
-  --eval.n_episodes 50 \
-  --eval.batch_size 8 \
-  --env.type aloha \
-  --env.task AlohaInsertion-v0
-```
-
------
+<details>
+<summary>📄 <strong>Click to view full <code>diffusion_aloha.yaml</code> used for training</strong></summary>
 
-## Training Configuration (Reference)
+```yaml
+# @package _global_
 
-For reproducibility, here are the key parameters used during the training session:
+# 随机种子
+seed: 100000
+job_name: Diffusion-Aloha-Insertion
 
-  - **Batch Size**: 8 (Limited by 8GB VRAM)
-  - **Optimizer**: Adam (lr=1e-4, betas=[0.95, 0.999])
-  - **Scheduler**: Cosine with warmup (500 steps)
-  - **Vision**: ResNet18 with GroupNorm
-  - **Precision**: Mixed Precision (AMP) enabled
+# 训练参数
+steps: 200000            # 原文件写的是 20万步 (Aloha 比较难练)
+eval_freq: 20000         # 稍微改频一点，方便看进度
+save_freq: 20000
+log_freq: 200
+batch_size: 8            # ⚠️ 关键：Aloha 必须用小 Batch，否则 8G 显存不够
 
-<!-- end list -->
+# 数据集
+dataset:
+  repo_id: lerobot/aloha_sim_insertion_human
 
-```bash
-# Original Training Command
-python -m lerobot.scripts.lerobot_train \
-  --config_path diffusion_aloha.yaml \
-  --env.type aloha \
-  --env.task AlohaInsertion-v0 \
-  --dataset.repo_id lerobot/aloha_sim_insertion_human \
-  --wandb.enable true \
-  --job_name DP_Aloha_Insertion \
-  --policy.repo_id Lemon-03/DP_Aloha_Insertion_test
-```
+# 评估设置
+eval:
+  n_episodes: 50
+  batch_size: 8          # 保持与训练一致
 
-### Config File (`diffusion_aloha.yaml`) Snippet:
+# 环境设置
+env:
+  type: aloha
+  task: AlohaInsertion-v0
+  fps: 50
 
-```yaml
-# Training
-steps: 200000
-eval_freq: 20000
-save_freq: 20000
-batch_size: 8
-
-# Policy
+# 策略配置
 policy:
   type: diffusion
+
+  # --- 视觉处理 ---
   vision_backbone: resnet18
+  # Aloha 的图片是矩形的，这里使用特定的裁剪尺寸
   crop_shape: [420, 560]
+  crop_is_random: true
+  pretrained_backbone_weights: null  # 原配置指定不加载预训练权重
+  use_group_norm: true
+  spatial_softmax_num_keypoints: 32
+
+  # --- Diffusion 核心架构 (U-Net) ---
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  diffusion_step_embed_dim: 128
+  use_film_scale_modulation: true
+
+  # --- 动作预测参数 ---
   n_action_steps: 8
   n_obs_steps: 2
   horizon: 16
+
+  # --- 噪声调度器 (DDPM) ---
+  noise_scheduler_type: DDPM
+  num_train_timesteps: 100
   num_inference_steps: 100
+  beta_schedule: squaredcos_cap_v2
+  beta_start: 0.0001
+  beta_end: 0.02
+  prediction_type: epsilon
+  clip_sample: true
+  clip_sample_range: 1.0
+
+  # --- 优化器 ---
+  optimizer_lr: 1e-4
+  optimizer_weight_decay: 1e-6
+  #grad_clip_norm: 10
+  
+  scheduler_name: cosine
+  scheduler_warmup_steps: 500
+
+  use_amp: true
 ```
+````
+
+\</details\>
+
+-----
 
+## 🚀 Evaluate (My Evaluation Mode)
+
+Run the following command in your terminal to evaluate the model for 50 episodes and save the visualization videos:
+
+```bash
+python -m lerobot.scripts.lerobot_eval \
+  --policy.type diffusion \
+  --policy.pretrained_path Lemon-03/DP_Aloha_Insertion_test \
+  --eval.n_episodes 50 \
+  --eval.batch_size 8 \
+  --env.type aloha \
+  --env.task AlohaInsertion-v0
 ```
 
----
+To evaluate this model locally, run the following command:
+
+```bash
+python -m lerobot.scripts.lerobot_eval \
+  --policy.type diffusion \
+  --policy.pretrained_path Lemon-03/DP_Aloha_Insertion_test \
+  --eval.n_episodes 50 \
+  --eval.batch_size 8 \
+  --env.type aloha \
+  --env.task AlohaInsertion-v0
+```