Logic
63 lines
1.9 KiB
YAML
63 lines
1.9 KiB
YAML
# @package agent
|
||
defaults:
|
||
- /backbone@vision_backbone: resnet_diffusion
|
||
- /modules@state_encoder: identity_state_encoder
|
||
- /modules@action_encoder: identity_action_encoder
|
||
- /head: transformer1d
|
||
- _self_
|
||
|
||
_target_: roboimi.vla.agent.VLAAgent
|
||
|
||
# ====================
|
||
# 模型维度配置
|
||
# ====================
|
||
action_dim: 16 # 动作维度（机器人关节数）
|
||
obs_dim: 16 # 本体感知维度（关节位置）
|
||
|
||
# ====================
|
||
# 归一化配置
|
||
# ====================
|
||
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
|
||
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）
|
||
|
||
# ====================
|
||
# 相机配置
|
||
# ====================
|
||
camera_names: ${data.camera_names} # 条件相机顺序固定为 r_vis, top, front
|
||
num_cams: 3 # 摄像头数量 (r_vis, top, front)
|
||
|
||
vision_backbone:
|
||
num_cameras: ${agent.num_cams}
|
||
camera_names: ${agent.camera_names}
|
||
|
||
# ====================
|
||
# 扩散过程配置
|
||
# ====================
|
||
diffusion_steps: 100 # 扩散训练步数（DDPM）
|
||
inference_steps: 10 # 推理时的去噪步数（DDIM，��定为 10）
|
||
|
||
# ====================
|
||
# Head 类型标识（用于VLAAgent选择调用方式）
|
||
# ====================
|
||
head_type: "transformer" # "unet" 或 "transformer"
|
||
|
||
# Head 参数覆盖
|
||
head:
|
||
input_dim: ${agent.action_dim}
|
||
output_dim: ${agent.action_dim}
|
||
horizon: ${agent.pred_horizon}
|
||
n_obs_steps: ${agent.obs_horizon}
|
||
# Transformer的cond_dim是每步的维度
|
||
# ResNet18 + SpatialSoftmax(32 keypoints) = 64维/相机
|
||
# 计算方式：单相机特征(64) * 相机数(3) + obs_dim(16) = 208
|
||
cond_dim: 208
|
||
causal_attn: false
|
||
time_as_cond: true
|
||
obs_as_cond: true
|