refactor: retire offpolicy go1/go2 joystick configs

conf/offpolicy/config.yaml

@@ -1,10 +1,10 @@
 defaults:
   - _self_
   - algo: sac
-  - task: ${algo}/go1_joystick_flat/mujoco
+  - task: ${algo}/g1_walk_flat/mujoco
 
 training:
-  task_name: Go1JoystickFlat
+  task_name: G1WalkFlat
   device: null
   logger: tensorboard
   wandb_project: unilab

conf/offpolicy/task/td3/g1_walk_flat/mujoco.yaml

@@ -0,0 +1,37 @@
+# @package _global_
+training:
+  task_name: G1WalkFlat
+  sim_backend: mujoco
+algo:
+  max_iterations: 100000
+env:
+  control_config:
+    action_scale: 1.0
+  gait_phase_init_mode: "offset_phase"
+  reset_base_qvel_limit: 0.5
+  noise_config:
+    scale_gyro: 0.0
+    scale_gravity: 0.0
+    scale_joint_angle: 0.01
+    scale_joint_vel: 0.1
+    scale_linvel: 0.0
+reward:
+  scales:
+    tracking_lin_vel: 2.0
+    tracking_ang_vel: 1.5
+    penalty_ang_vel_xy: -1.0
+    penalty_orientation: -10.0
+    penalty_action_rate: -4.0
+    pose: -0.5
+    penalty_feet_ori: -20.0
+    feet_phase: 5.0
+    alive: 10.0
+  tracking_sigma: 0.25
+  base_height_target: 0.754
+  min_base_height: 0.3
+  max_tilt_deg: 65.0
+  gait_frequency: 1.5
+  feet_phase_swing_height: 0.09
+  feet_phase_tracking_sigma: 0.04
+  close_feet_threshold: 0.15
+  pose_weights: [0.01, 1.0, 5.0, 0.01, 5.0, 5.0, 0.01, 1.0, 5.0, 0.01, 5.0, 5.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0, 50.0]

docs/users/zh_CN/01-getting-started.md

@@ -68,7 +68,7 @@ uv run scripts/train_appo.py task=go1_joystick_flat/mujoco
 
 # SAC / TD3
 uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco
 ```
 
 ### 验证环境

docs/users/zh_CN/02-simulation-backends.md

@@ -65,12 +65,9 @@ uv run scripts/generate_support_matrix.py --write
 | APPO (torch) | `g1_motion_tracking` (G1 motion tracking) | Tested | Tested |
 | APPO (torch) | `g1_flip_tracking` (G1 flip tracking) | Tested | Tested |
 | APPO (torch) | `allegro_inhand` (Allegro in-hand) | Tested | Registered |
-| SAC (torch) | `go1_joystick_flat` (Go1 joystick) | Tested | Tested |
-| SAC (torch) | `go2_joystick_flat` (Go2 joystick) | Tested | Tested |
 | SAC (torch) | `g1_walk_flat` (G1 walk flat) | Tested | Tested |
 | SAC (torch) | `g1_walk_rough` (G1 walk rough) | Tested | Registered |
-| TD3 (torch) | `go1_joystick_flat` (Go1 joystick) | Tested | Tested |
-| TD3 (torch) | `go2_joystick_flat` (Go2 joystick) | Tested | Tested |
+| TD3 (torch) | `g1_walk_flat` (G1 walk flat) | Tested | Registered |
 
 ### Source Index
 

docs/users/zh_CN/03-training.md

@@ -28,8 +28,8 @@ uv run scripts/train_mlx_ppo.py task=go1_joystick_flat/mujoco
 uv run scripts/train_appo.py task=go1_joystick_flat/mujoco
 
 # Off-policy
-uv run scripts/train_offpolicy.py algo=sac task=sac/go1_joystick_flat/mujoco
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco
 
 # CLI override
 uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco algo.num_envs=2048 algo.max_iterations=1000
@@ -50,20 +50,20 @@ run 目录命名格式是 `YYYY-MM-DD_HH-MM-SS_<sim_backend>`，例如 `2026-03-
 ```bash
 # 回放最新结果
 uv run scripts/train_rsl_rl.py task=go2_joystick_flat/mujoco training.play_only=true
-uv run scripts/train_offpolicy.py algo=sac task=sac/go2_joystick_flat/mujoco training.play_only=true
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco training.play_only=true
 
 # macOS / MacBook 上的 MotrixSim 原生 renderer
 uv run mxpython scripts/train_rsl_rl.py task=go2_joystick_flat/motrix training.play_only=true
 
 # 回放指定 run
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco training.play_only=true algo.load_run="2024-02-04_12-00-00"
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco training.play_only=true algo.load_run="2024-02-04_12-00-00"
 ```
 
 ## Resume Training
 
 ```bash
 uv run scripts/train_rsl_rl.py task=go2_joystick_flat/mujoco algo.load_run="2024-02-04_12-00-00"
-uv run scripts/train_offpolicy.py algo=sac task=sac/go2_joystick_flat/mujoco algo.load_run="2024-02-04_12-00-00"
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco algo.load_run="2024-02-04_12-00-00"
 ```
 
 ## Hydra Overrides
@@ -116,10 +116,10 @@ uv run scripts/train_appo.py \
 # 按 task 分组
 uv run scripts/train_offpolicy.py \
   algo=sac \
-  task=sac/go2_joystick_flat/mujoco \
+  task=sac/g1_walk_flat/mujoco \
   training.logger=wandb \
   training.wandb_project=unilab-benchmark \
-  training.wandb_group=go2_joystick_flat
+  training.wandb_group=g1_walk_flat
 ```
 
 常用字段:

docs/users/zh_CN/04-algorithms.md

@@ -84,29 +84,29 @@ FastSAC 和 FastTD3 使用同一套异步多进程架构，通过 shared memory
 
 ```bash
 # 基本训练
-uv run scripts/train_offpolicy.py algo=sac task=sac/go2_joystick_flat/mujoco
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco
 
 # 异步采集模式
-uv run scripts/train_offpolicy.py algo=sac task=sac/go2_joystick_flat/mujoco training.no_sync_collection=true
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco training.no_sync_collection=true
 
 # 跳过自动回放
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco training.no_play=true
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco training.no_play=true
 ```
 
 ### Playback
 
 ```bash
-uv run scripts/train_offpolicy.py algo=sac task=sac/go2_joystick_flat/mujoco training.play_only=true
-uv run scripts/train_offpolicy.py algo=td3 task=td3/go1_joystick_flat/mujoco training.play_only=true algo.load_run="2024-02-04_12-00-00"
+uv run scripts/train_offpolicy.py algo=sac task=sac/g1_walk_flat/mujoco training.play_only=true
+uv run scripts/train_offpolicy.py algo=td3 task=td3/g1_walk_flat/mujoco training.play_only=true algo.load_run="2024-02-04_12-00-00"
 ```
 
 ### Key Parameters
 
 | 参数 | 默认值 | 说明 |
 |------|--------|------|
 | `algo` | `sac` | 算法选择 |
-| `task` | `sac/go1_joystick_flat/mujoco` | 单个 task 配置入口，内部同时定义 algo + task + backend |
+| `task` | `sac/g1_walk_flat/mujoco` | 单个 task 配置入口，内部同时定义 algo + task + backend |
 | `algo.max_iterations` | 500 (SAC) / 5000 (TD3) | 最大训练迭代数 |
 | `algo.num_envs` | 4096 | 并行环境数量 |
 | `training.device` | 自动检测 | learner 设备 |